SYSTEM
STABILITY
M. A. I'Al
Prufwsur Indian Inxl.ihir.fTinWhnolnpy Kii!iinr, Imlia Wilinj; I'rnfpKrtftr luwii
twUonM.
I.
Htiu.
ii. Sn4.
AACR2
1SKN a li:ic*13 0
Research work in applying Lyapunov Stability theory to power systems has been pursued actively for the last decade and a half. The amount of literature is quite vast and there is now strong evidence that the method may find eventual application in planning and online dynamic security assessment. It was, therefore, felt that it is now appropriate to collate, unify and present in a cohesive conceptual framework both the theory and application aspects of the direct method of stability analysis for power systems. After a physical statement of the problem in the introductory chapter, Lyapunov's stability theory is reviewed in Chapter Two. We develop the mathematical models for multimachine power systems in Chapter Three stressing both the center of angle formulation and the Lure formulation of the equations. The systematic construction of Lyapunov functions for single and multimachine power systems is the theme of Chapter Four. It is shown that the energy function is one of the many Lyapunov functions that can be constructed. The computation of the region of attraction of the postfault The recent approaches which make use of the stable equilibrium point is the most difficult problem in Lyapunov stability analysis of power systems. faulted dynamics of the system are emphasized in Chapter Five as opposed to earlier results which gave conservative estimates of the critical clearing time. Two case studies are presented. In Chapter Six we discuss the (often termed as Both the weighted sum scalar stability of large scale power systems by decomposition the vector Lyapunov function approach). function approaches are discussed.
Lyapunov function and the comparison principlebased vector Lyapunov The final chapter concludes with a
PREFACE
survey of some recent results in this rather active area of research and some thoughts on problems still requiring a solution. derived In the Appendix we have
PREFACE
the detailed and simplified models of the synchronous machine. The book is aimed at a research audience as well as an application oriented power engineer who needs to understand the basis of direct methods of stability analysis in power systems. The book can be used in a graduate level course in power system stability or dynamics. In writing this book, particular, I have been influenced considerably by the work of In "Transient Energy Stability Emergency N. Michel
many other research workers in this field whose works are cited. I would like to mention the report Analysis" by Systems Control Inc. Foundation Conference on Operating State Control' Switzerland in Oct. of 1979. in Proceedings of the Engineering Dept.
of Energy at Davos,
of Iowa State University for numerous interesting discussions on the stability large scale dynamical systems.
I would like to thank my former graduate students at I.I.T. Kanpur who got excited by this area of research and worked with great enthusiasm. gratitude to I.I.T. academic excellence and faculty support for research. and Technology, Kanpur. The writing of this monograph was completed during my stay at Iowa State University during 197981. Kopplin, Chairman, Dept. I am very grateful to Professor J. of Electrical Engineering, 0. for his constant I would Govt, My Kanpur is an eternal one for nurturing an atmosphere of I would like to thank of Science
like to thank Shellie Siders for her speedy and excellent job of typing the
Writing of this book has taken long hours away from my family, especially my son, Gurudutt, whom I thank for his understanding. M. A. Pai June 1981
CHAPTER
INTRODUCTION
1.1
Introduction
Any physical system that is designed or operated to perform certain preassigned tasks in a steady state mode must, these functions in a satisfactory manner, disturbances with an adequate margin of safety. in addition to performing When the physical system be stable at all times for sudden
is large and complex such as a typical modern interconnected power system, investigation of stability requires both analytical sophistication in terms of techniques employed and practical experience in interpreting the results properly. In the last decade and a half and in particular after the famous blackout in N.E. U.S.A. in 1965, considerable research effort has gone into the stability investigation of power systems both for offline and online purposes. At the design stage although the planner takes many in subsequent operation and augmentation contingencies into consideration, planner.
of the network, new considerations arise which were not foreseen by the Hence, an entirely different pattern of system behavior can be expected under actual operation conditions. This is particularly true regarding the capability of the system to maintain synchronism or stability due to sudden unforeseen disturbances such as loss of a major transmission line, load or generation. The tools suitable for offline studies such A technique which The appeal of this as simulation may not be suitable for online application since a large number of contingencies have to be simulated in a short time. offers promise for this purpose is Lyapunov1s method.
clearing time of circuit breakers for various faults and thus directly assess the degree of stability for a given configuration and operating state. disturbances that the system can withstand, security assessment. The critical clearing time can also be translated in terms of additional power thus offering a tool for dynamic the method can serve as For offline applications,
a complementary role to simulation by selecting quickly the important faults which need to be studied in detail. The state of art in this field resulting from nearly two decades of research work has reached a certain degree of maturity and for the first time offers possibilities of application by the
electric utilities.
method and its systematic application to power system stability. Perhaps for no other physical system has Lyapunov*s stability theory been applied so extensively and vigorously as for power systems. It is Since one difficult to precisely pinpoint historically the first application of Lyapunov's method for the investigation of power system stability. Lyapunov's function itself is a generalization of the energy function, system problem when the equal area criterion was first proposed. a power system was by Magnusson in 1947 [1].
can say that Lyapunov's method indeed was applied implicitly to the power The earliest work on applying the energy function to determine the stability of An energy integral criterion [2]. and as applied for multimachine system was proposed by Aylett in 1958 as a solution to the power system stability problem by Gless ElAbiad and Nagappan [4]. In Ref. [4], [3]
Lyapunov's method as we know it in control literature today was first proposed the method was put on a firm
footing as a powerful tool for stability analysis by power engineers through a realistic example of computing the critical clearing time and giving an algorithmic procedure for computer implementation. such as those by Willems Yu and Vongasuria [8], [5] , Fallside and Patel [9] and Siddiqee [10] The early research [6], Dharma Rao [7], Di
concentrated on applying
different methods of constructing Lyapunov functions for power systems. Caprio and Saccomanno and Luders [11]
related the physical aspects of the problem to the In 1970, [13], the systematic construction of Lyapunov Pai, Mohan and Rao [14] and Pai
mathematical model and constructed an energy type Lyapunov function for multimachine systems. [12], Willems and Willems [15]. time, system. function to power systems using Popov's criterion was proposed by Willems It was soon realized that the more important aspect of applying the namely the computation of critical clearing lay in computing accurately the region of attraction of the postfault This has received considerable attention by RibbensPavella [16,17]. Since 197C, there has been
an explosive growth in the research work on Lyapunov stability analysis of A number of stateoftheart papers have appeared since [18], RibbensPavella [19], [21] summarize thoroughly the research [22], the Hence, In particular, the papers by Willems [20] and Fouad
method still had its drawbacks in terms of not giving consistently good results for critical clearing time for a wide range of fault conditions. largely theoretical aspects of the problem were pursued by the research workers in the 70's. Recently a significant breakthrough in obtaining excellent practical results has been achieved by Athay, Podmore and Virmani
in U.S.A.
[23],
[24]
of its conservative nature of results is today gaining acceptance both as an offline tool for planning purposes as well as for online security assessment schemes. An alternative method of investigating stability of large scale The [26] and extended power systems is through decomposition and subsequent aggregation. concept of vector Lyapunov function proposed by Bailey to a wide class of composite systems by Michel and Porter Miller [28], and Siljak [31]. [29,30] Pai and Narayana
Improvements in decomposition techniques in terms of reducing the number of subsystems as well as handling transfer conductances were proposed by Jocic, RibbensPavella and Siljak [32]. viewpoint are still conservative. However, results from a practical this promises to be an There are two other In the future,
area of active research and any breakthrough will have a major impact on the stability analysis of large scale power systems. promising research areas. the topology of the transmission network [33] One consists in stability analysis by retaining thus offering the possiblity The other direction in [35,36].
of a better understanding of the phenomenon of loss of synchronism and also accomodating nonlinear load representation synchronous machines, 1.2 excitation systems, [34]. etc. which research is being pursued is incorporating accurate models of
The intuitive idea of stability of a physical system is as follows: the system be in some equilibrium state. disturbance, say the system is stable. equilibrium point. unstable.
If on the occurrence of a
the system eventually returns to the equilibrium position, we The system is also termed stable if it converges so that certain
to another equilibrium position generally in the proximity of the initial If the state of the system "runs away" physical variables go on increasing as t * ~, then we say the system is These intuitive ideas are also applicable to a power system. the power system is operating in a synchronous i.e. the rotor angles of the The constant nature of
different machines with respect to a rotating synchronous reference frame are fixed and the system frequency is constant. of the system is not accelerating. equilibrium. frequency implies that the system as a whole is at rest or the inertial center The occurrence of a disturbance tends to momentarily alter the synchronous equilibrium and the frequency The disturbance may either be small or big and the system may become unstable in
either event depending on the operating condition at t = t . The study of stability in the presence of small disturbances constitutes what is known as "static stability" or differential equations. "dynamic stability" analysis in the literature. the The mathematical model for such a study is a set of linear timeinvariant When the disturbances are large, nonlinearities inherent in the power system can no longer be ignored and the study of stability under such circumstances constitutes what is known as "transient stability" analysis. The mathematical model for such a study is a set of nonlinear differential equations coupled with a set of nonlinear algebraic equations. The dimensionality of the mathematical model both for dynamic and transient stability can be very large, systems. and now look at the physical phenomenon. Assume that the system is in steady state with a frequency to and the rotor angles 6? of the various machines with respect to the synchronously rotating reference frame (i.e., 6^ = a.  a) t where a . is the angle with respect to a fixed 3 E 1 o 1 reference) assuming constant values. Under these conditions, the power At t = tQ, supplied by the generators exactly matches the power absorbed by the loads plus the electric power loss in the transmission system. line or a fault at one of the buses. suppose a large perturbation in the system occurs, such as the loss of a major This disturbance upsets the energy For example, a fault on a This results balance existing prior to the disturbance. for even moderately sized We shall restrict ourselves to the transient stability aspects
transmission line blocks the transfer of power to loads from certain nearby generators and will try to absorb power from other generators. in an excess or deficit of mechanical power supplied over the electrical power at each generator in the system resulting in acceleration or deceleration of the rotors. The rotor angles 6^ with respect to the synchronously rotating A 1.1. When the fault reference frame U)q either increase or decrease as a function of time. typical behavior or rotor angles is shown in Fig. possible, we have the problem of the redistribution of the excess kinetic energy acquired by the system until the instant of fault clearing. absorb this energy, If the system after the fault is cleared can then instability results. 1.2 an Knowledge it may recover the system is considered to be transiently stable; if
Figure 1.1 shows a system which is transiently stable and Fig. When the system is transiently stable, to its original frequency or settle down to a new frequency. planner and the operation engineer. in instability.
regarding transient stability of the system is important both to the system The system planner has to coordinate the relay settings so that a given fault is cleared in time so as not to result A valuable piece of information in
10
of circuit cr breakers which isolates the faulty portion from the rest of the system. A fault cleared before t will result in a J cr stable system and a fault cleared after t will result in the 1 cr loss of synchronism. In order to compute ^  c r t the system planner will have to conduct several stability runs each time fixing the clearing time If the system is stable, t , J is repeated. and simulating the entire system. is increased and the simulation cl = t , an This is a tedious Considering the . This is
inspection of the swing curves reveals instability. process besides being expensive in terms of computer time.
fact that this procedure has to be repeated for different faults in the system, the need is evident for some direct method of computing t
C IT
precisely sought to be done by Lyapunov's method. In online applications, it is necessary to convert the information about For example, the difference between the It is also possible to consider the The application of
the critical clearing time into some kind of a security index for interpretation by the operator. transient security index [37]. actual clearing time and the critical clearing time can be converted into a effects of the outages on the transient stability. of research [38,39,40].
11
In Lyapunov's method, a certain scalar function of system state variables known as Lyapunov function V(x) analytically. In many cases, enough candidate for V(x). denoted by the expression V(x) is computed first the system energy is a good
12
c r is
In terms of energy,
the critical energy of the system which if exceeded during the the system will not be able to absorb it There are many ways of computing and more the
computation of Vcj. is made fault dependent instead of being valid for all faults in the system for a given postfault configuration. considerable research effort has been directed towards constructing better Lyapunov functions, accommodating accurate models of synchronous machines and extending the methodology for large scale systems through decomposition and vector Lyapunov function approach. All the collective research work done As a planning tool, Lyapunov's so far holds promise for Lyapunov's method being used as a tool for planning as well as online security assessment. t stability analysis can be performed on a large number of faults to compute and consider for detailed study only those which
J
the can be cl
at t = t , 
converted into stability margins for different contingencies and have the information displayed to the operator in a meaningful manner. 1.3 Chapter outline
We first give a basic review of Lyapunov stability theory in Chapter II. Various methods of constructing Lyapunov functions are covered. an independent and selfcontained treatment of the subject. power systems. Lyapunov It is then stability theory is discussed in slightly greater detail so as to serve as followed in Chapter III by developing the mathematical model of multimachine Both the center of angle reference frame and machine angle The reference models are developed.
13
systematic construction of Lyapunov function for both single and multimachine models is then emphasized in Chapter IV using control theoretic as well as physical reasoning. For the single machine case, the region of attraction is computed. Computation of the region of stability for the multimachine case is the single biggest impediment to the method being used for online purposes. This is discussed in Chapter V. The transient energy function approach which takes into consideration the effect of faulted dynamics in the computation of V"cr and which has resulted in the accurate computation of t is stressed. In Chapter VI, the vector Lyapunov In the final chapter, function approach via decomposition is discussed. analysis by Lyapunov's method are briefly reviewed.
1C
POWER
References 1. 2.
3.
4.
5.
6.
7.
8.
Yu, Y. N. and Vong Stability Study by Lya PAS, Vol. 86, No.
9.
10.
11.
12.
13.
14.
15.
16.
RibbensPavella, M. "Transient Stability of Multimachine Power Systems by Lyapunov's Method", IEEE Winter Power Meeting, New York, 1971. RibbensPavella, M. and Lemal, B., "Fast Determination of Stability Regions for Online Transient Power System Studies",
17.
INTRODUCTION
15
(London), Vol.
123,
No.
7, July 1976,
pp.
Willems, J. L., "Direct Methods for Transient Stability Studies in Power System Analysis", IEEE Trans. Automatic Control, Vol. A616, No. 4, Aug. 1971, pp. 332341. RibbensPavella, M., "Critical Survey of Transient Stability Studies of MultiMachine Power Systems by Lyapunov's Direct Method", Proceedings of 9th Annual Allerton Conference on Circuits and System Theory, Oct. 1971. Saccomanno, F., "Global Methods of Stability Assessment", Proceedings of Symposium on Power System Dynamics, The University of Manchester Institute of Science and Technology, Manchester, England, Sept. 1973. Fouad, A. A., "Stability TheoryCriteria for Transient Stability", Proc. Engineering Foundation Conference on "System Engineering for Power", Henniker, New Hampshire, 1975, pp. 421450. Willems, H. F., Louie, S. A. and Bills, G. W., "Feasibility of Lyapunov Functions for the Stability Analysis of Electric Power Systems Having up to 60 Generators", IEEE Trans. PAS, Vol. 91, No. 3, May/June 1972, pp. 11451153. Athay, T., Podmore, R. and Virmani, S., "A Practical Method for Direct Analysis of Transient Stability", IEEE Trans. PAS, Vol. 98, No. 2, March/April 1979, pp. 573584. RibbensPavella, M., Gruijc, Lj T., Sabatel, J. and Bouffioux, A., "Direct Methods for Stability Analysis of Large Scale Power Systems", Proc. IFAC Symposium on "Computer Applications in Large Scale Power Systems", New Delhi, India, Aug. 1618, 1979, Pergamon Press, U.K. Kakimoto, N., Ohsawa, Y., and Hayashi, M., "Transient Stability Analysis of Electric Power System vis Lure"* Type Lyapunov Function", Parts I and II, Trans. IEE of Japan, Vol. 98, No. 5/6, May/June, 1978. Bailey, F. N., "The Application of Lyapunov's Method to Interconnected Systems", Journal SIAM (Control) Ser. A., Vol. 3, No. 3, 1966, pp. 443462. Michel, A. N. and Porter, D. W., "Stability of Composite Systems", Proc. Fourth Asilomar Conference on Circuits and Systems, Monterey, California, 1970. Michel, A. N. and Miller, R. K., "Qualitative Analysis of Large Scale Dynamic Systems" (Book), Academic Press, New York, 1977. Siljak, D. D., "Stability of Large Scale Systems", Proc. 5th IFAC World Congress, Paris, 1972. Siljak, D. D., "Large Scale Dynamic Systems  Stability and Structure" (Book) North Holland, New York, 1978. Pai, M. A. and Narayana, C. L., "Stability of Large Scale Power Systems", Proc. 6th IFAC World Congress, Boston, 1975. Jocic, Lj B., RibbensPavella, M., and Siljak, D. D., "On Transient Stability of MultiMachine Power Systems", IEEE Trans. Automatic Control, Vol. AC23, 1978, pp. 325332. Bergen, A. R. and Hill, D. J., "A Structure Preserving Model for Power Systems Stability Analysis", IEEE Trans. PAS, Vol. ICG, Jan. 1981, pp. 2535.
19.
2C.
21.
22.
23.
24.
25.
26.
27.
28.
33.
INTRODUCTION
16
34.
Athay, T. M. and Sun, D. I., "An Improved Energy Function for Transient Stability Analysis", Proc. International Symposium on Circuits and Systems, April 2729, 1981, Chicago. Sasaki, H., "An Approximate Incorporation of Field Flux Decay into Transient Stability Analysis of Multimachine Systems by the Second Method of Lyapunov", IEEE Trans. PAS, Vol. 98, No. 2, March/April 1979, pp. 473483. Kakimoto, N., Ohsawa, Y. and Hayashi, M., "Transient Stability Analysis of Multimachine Power Systems with Field Flux Decays via Lyapunov's Direct Method", IEEE Trans. PAS, Vol. 99, No. 5, Sept./Oct. 1980, pp. 18191827. Teichgraeber, R. D., Harris, F. W., and Johnson, G. L., "New Stability Measure for Multimachine Power Systems", IEEE Trans. PAS89, Vol. 2, Feb. 1970, pp. 233239. RibbensPavella, M., Calavaer, A. and Ghewry, A., "Transient Stability Index for Online Evaluation", IEEE PES Winter Power Meeting, 1980. Saito, O., Kinnosuke, K., Muneyuky, U., Masahiro, S., Hisao, M. and Toshihiko, T., "Security Monitoring System Including Fast Transient Stability Studies", IEEE Trans., PAS94, Sept./Oct. 1975, pp. 17891805. Fouad, A. A., Stanton, S. E., Mamandur, K.R.C., Kruempel, K. C. , "Contingency Analysis Using the Transient Energy Margin Technique", Paper 81 SM 3979, IEEEPES Summer Power Meeting, Portland, Oregon, July 2631, 1981.
35.
36.
37.
38.
39.
40.
CHAPTER II
2.1
Introduction
The theory of stability of linear timeinvariant systems is well known in the literature through the methods of Nyquist, RouthHurwitz, etc. provide both necessary and sufficient conditions. However, of nonlinear systems, no such systematic procedures exist. rule. which in the case Closedform
solutions of nonlinear differential equations are exceptions rather than the Although the digital computers can numerically integrate the they do not provide an insight into the qualitative Consequently, the problem of differential equations,
finding the necessary and sufficient conditions for the stability of nonlinear systems is a formidable one and as yet an unsolved problem. All efforts in this direction relate directly or indirectly to the work of A.
M.
Lyapunov
[1]
who in 1892 set forth the general framework for the Lyapunov dealt with both linear and nonlinear
He outlined two approaches to the problem of stability, known The distinction is based on the fact no such knowledge is
popularly as Lyapunov's "first method" and the other as the "second method of Lyapunov" or the "direct" method. differential equations. necessary. that the "first method" depends on finding approximate solutions to the In the "second method", This is a great advantage in the case of nonlinear systems.
We will be mainly concerned with the "second method of Lyapunov" or the "direct method" as applied to nonlinear systems. A second major breakthrough in nonlinear stability theory came through the work of V. M. Popov [2] who in 1962 outlined a
criterion in the frequency domain for certain classes of nonlinear control systems. This method is particularly attractive because of the ease with which it can be applied in much the same way as Nyquist's criterion is applied to linear systems. The mathematical model of power systems is amenable to analysis by Popov's method under certain simplifying assumptions. A literature survey of the area of nonlinear stability using Lyapunov's method is practically impossible because of the vast amount of research surrounding this topic. (Hahn [3], There are a number of texts devoted to this topic constituting LaSalle and Lefschetz [4], Willems [5], Vidyasagar [6]). a good source of reference material on various aspects of stability theory What follows is a selfcontained treatment of the topic with an emphasis on applications. 2.2 Mathematical Description of Systems
Most physical systems are described by their mathematical models for purposes of analysis. The following are some of the descriptions that are possible u in continuous time systems, x represents the n dimensional state vector, time variable. 1 2 3 4 x = f x = f k = f x = f (x,u,t) (x,t) (x,u) (x) Nonlinear, Nonlinear, Nonlinear, time varying and forced
The mathematical model of power systems for stability analysis results in an autonomous system of type (4) and fortunately for such type of systems a number of systematic procedures exist for constructing Lyapunov functions. In this chapter, we restrict the treatment to autonomous systems.
17
II
15
(2.1 ) i.e. *2 x
= f (x
= ^ Xj^ 2 l'x2V
2* * * " '
n s Let x
2' (2.1),
' rr i.e.
be a solution of
xs = f(xs) (2.2 ) If xg = constant, C = f(xg) (2.3 ) Let at some instant of time, which can be designated as t = C without loss of generality, the system be disturbed instantaneously to a position x^ in The motion of the system for t > 0 is denoted by x(t). . Hence (2.1) becomes in view of (2.2) (2.4) the state space. Define y = x  x then
 f(x )
we may
=
9(y)
This
is often referred to as change of coordinates or transferring the equilibrium solution to the origin. Henceforth, we shall assume that this has been done and consider the stability of the origin equations x = f(x), f(0) = 0 (2.5) (equilibrium point) of the
II
15
The origin is equivalently called the trivial solution, singular point or the equilibrium point. the origin is the only equilibrium point of (2.5). Furthermore, we will assume that (2.5) (2.5) For the sake of simplicity, let us assume that
satisfies the socalled Lipschitz condition so that the solution of exists and is unique for a given initial condition x^. 2.3 Definitions of Stability
We are all familiar with the intuitive idea of stability discussed in Chapter I. But precise definitions are needed before we can formulate the problem in more quantitative terms. In the mathematical literature on this subject, there are scores of definitions each differing from the other in a subtle way. However, for engineering purposes, we need to consider only about the stability of a certain motion of the system (2.1). An undisturbed motion x of s (2.1) satisfying (2.2) is con
sidered to be stable if by giving a small disturbance to it, the disturbed motion remains close to the unperturbed one for all time. specifically: a) If for small disturbances, the effect on the motion is small, one says that the undisturbed motion is 'stable'. b) If for small disturbances, the effect is considerable, the undisturbed motion is termed 'unstable'. c) If for small disturbances, the effect tends to disappear, the undisturbed motion is 'asymptotically stable'. d) If regardless of the magnitude of the disturbances, the effect tends to disappear, the undisturbed motion is 'asymptotically stable in the large'. The above qualitative discussion will be put in more quantitative terms. Consider the autonomous system x = f(x) (2.6) More
where the origin representing the undisturbed motion is the equilibrium point in the state space, 0 = f(0) Stability i.e.
2.3.1
Definition 1: The origin is said to be stable in the sense of Lyapunov or simply stable if for every real number e > 0 and initial time t. exists a real number 6 > 0
u
> 0 there
depending on e and in general on t such that for all initial conditions satisfying the inequality,
IIs0 El < *
II
15
 x (t) 
< e
The symbol  [ stands for the norm. In the Euclidean space, the norm can be interpreted to mean the Euclidean norm, n ( I
5 1/2 x 1 x/) i=l
i.e.,
= It does
not tell us apriori how small 6 has to be chosen in the definition. The origin is termed unstable if for a given e as above and for sufficiently small 6 there exists one solution which does not satisfy the inequality, x(t)  < e.
The above definition is unsatisfactory from an engineering point of view since one is interested in a strong type of stability, namely that the solution must return to the origin as t > co. 2.3.2 Asymptotic Stability The origin is said to be asymptotically stable if it is
Definition 2:
stable and every motion starting sufficiently close to the origin converges to the origin as t tends to infinity, i.e. t lim x(t)  * 00
y
(a)
Stability
/ f
/>
'\
V
\ IV
II
(b)
Asymptotic stability
(c)
A deficiency of the above definition is that it does not indicate how big the disturbances can be in order that all subsequent motions will converge to the origin. 2.3.3 In that sense, it is again a local concept.
Asymptotic Stability in the Large The origin is said to be asymptotically stable in the large
Definition 3:
if it is asymptotically stable and every motion starting at any point in the state space, returns to the origin as t tends to infinity. This is a very desirable type of stability in engineering systems since one does not have to worry about the magnitude of disturbances. definitions can be illustrated geometrically as in Fig. twodimensional case. 2.3.4 Absolute Stability 2.2(a). The above three 2.1 for a
Consider the nonlinear control system with zero input shown in Fig. The nonlinearity is known to pass through the origin, and a line of given slope k. a sector condition.
and third quadrants and is restricted to a region between the horizontal axis Such a type of nonlinearity is said to satisfy If for the origin The linear transfer function G(s) is fixed.
is asymptotically stable in the large, we say that the nonlinear system is absolutely stable. For multiple nonlinearities, we can extend the concept to a transfer function matrix W(s) (Fig. 2.2(b)) with a vector
20
such f^(g)
= fV(#v)
and
each ^(o^)
In many physical situations, the origin may not be asymptotically stable for all possible initial conditions but only for initial conditions contained in a region around the origin. Such a region is called the region of If such a region asymptotic stability or finite region of attraction.
Ha)
(a) Single nonlinearity
f (o)
W(s)
(b)
22
exists, designer.
computation of such a region is of great interest to the system For example, in a multimachine power system, if a fault is
cleared too late, the postfault system may not return to a stable state. This can be interpreted equivalently as the computation of a region of attraction around the postfault stable state and determining if the state of the system at the point of clearing the fault lies within this region or not. Examples: i) Stability:
X
= X
~X1
The origin is stable since given an e within which we want the trajectory to lie, we can always give such an initial condition x^ lying within a circle of radius 6, ii) (6<e) such that the trajectory lies within e. Consider Vanderpol's equation 2 e > 0. The system has negative damping for Instability:
x + e (x l)x + x = 0,
small x and positive damping for large x. It has a stable limit cycle. The origin is unstable but the motion of the system is bounded, iii) Stability:
X
= X
*2
"Xl "
2 Region of
The origin is asymptotically stable in the large, iv) Asymptotic Stability: t = t x  e (x2l)x + x = 0
The origin is asymptotically stable but not in the large. is a finite region of asymptotic stability. however, unstable
There
since a small perturbation around the limit cycle will take the trajectory away from the limit cycle. 2.5 Basis of Lyapunov's Method
The "direct" method of Lyapunov represents a philosophy of approach to the problem of stability and yet it is a very basic concept. The principle idea of the method is contained dE 'If the rate of change ~ of of an isolated physical system is negative for
22
xg,
then the energy will continually decrease until it finally In our formulation x =0. e there
The above concept was developed into a precise mathematical tool by Lyapunov. When the description of the system is given in a mathematical form, is no way of defining energy. Hence, the energy function of E(x) by some scalar function V(x). a function V(x) is zero and its derivative V(x) function V(x) is replaced
If for a given system, one is able to find is < 0 except at x = 0 where it is zero, The In what follows, we shall Note that
then we say the system returns to the origin if it is disturbed. is called the Lyapunov function. state somewhat more precisely Lyapunov's theorem on stability. in the computation of V(x), shown below: Consider the autonomous equation f (0) = 0 3x_9V_^n dt ' '
x = f (x) , *M K '
8x
'
>
f (x)
>
will now be
Definition 4: A function V(x) is called positive (negative definite definite) if V(0) = 0 and if around the V(x> > 0 (<0) for origin x / 0. A function V(x) is called positive definite semi{negative semidcfiniLe) if V(0) = 0 and the origin V(x)  0 if around Definition (<0)
b :
tor x / 0.
For n = 3 2 V(x) +
(X2+x^J
24
V(x)
Stability Theorems
In this section, we state the theorems on stability of the origin of the autonomous equation defined by (2.6). 2.7.1 Theorem on Stability (2.6) is said to be stable if there exists a scalar <
ft2  xx The origin is the only equilibrium point as can be seen by inspection. Choose V(x) = xj + x., which is positive definite.
= 0 Hence, the condition V(x) Therefore, < 0 is satisfied and V(x) > 0
with the differential equations, namely, is easily verified that V(x) system.
conservative system and the total energy Example: Consider *i * Lhe system of equations
**  *3 3
i <*22*3)2
24
2x
2 " X3
V(x) is negative semidefinite, being zero for = = 0 and x^ / 0. origin is therefore stable. 2.7.2 Theorem on Asymptotic Stability (2.6) is asymptotical1y stable < 0.
The
it is called a Lyapunov function for the system or simply the V function. Example: Consider the system of equations
X
l 2
= X
2 "
aX
(x +x
2 i 2*
~*1 "
nx <X +X )
2 l 2
where
a?
is a positive constant. 2 2  x 1 + x2
Choose
i"flx2) t^xi+bx2"1J
(ax^+x^J (ax2+bx2l)
V(x) = 2fax2 + Bx2) (ax2+bx2l). If a, b, o, v(x) 2 U are all positive, then V(x) is positive definite and
24
ax' + bx
< i.
The origin is asymptotically stable for all initial Note that in this instance by examining V(x)
we are able to get an estimate of the domain of attraction. The exact region of asymptotic stability is
bigger than this since Lyapunov's theorem gives only sufficient but not necessary and sufficient conditions. 3.7.3 Theorem on Asymptotic Stability in the Large (2.6) is asymptotically stable in the large if there such that
The origin of
exists a scalar function V(x) i) iii) V(x) V(x) ' > 0 < 0
00
ii) V(x)
as
x
i.e.,
a
V(x) (C
Condition
(iii)
closed surfaces for all C in the entire state space. radially unbounded function is 2 2
V(x)
of a
= x^
1 2 = = 
Theorem 2.7.3 is difficult to apply in > 0 with V(x) Hence, we have an alternative theorem.
practical situations because it is difficult to find V(x) in the entire state space.
2.7.4
(ii)
(a) V(x)
<
does not vanish identically for t > 0 along any other (ii) implies that if V(x) = 0 at points other than the then these points should not constitute a
dT
1 / i d t
i f<i) + L
28
*2
= _
UC
l " E
UX
2}
1
+
L x
V(xlfx2)
L x2>c2
'"IT*
0 and V(x) < 0.
L X
" iZXl 
Z* 2
flX
2 "  "X2
f(X
2>
Assume positive values of resistance only so that f(x2) > 0. Hence, V(x) > Also V(x} + ~ as x * . the origin. We now invoke the alternative V(x) = 0 for x^ 0 but formulation of the theorem on a.s.i.l. and verify that V(x} does not vanish along any solution other than variable i.e. the x^ axis. But it does not constitute a solution of the  0. Therefore, we conclude that
In many physical systems, it is much easier to find V(x) > 0 and V(x) rather found to be most readily amenable for engineering applications. that V{x) from physical considerations as in the preceding example. 2.7.5 Theorem on Region of Asymptotic Stability
than a strict negative definite V{x). Hence, Theorem 2.7.4 will be The fact ^ 0 along any solution other than origin should be argued out
Consider the autonomous equation ft = f (x) , f (C)  0 Let f/ designate a region where V(x)
0 < 0
Then the origin is asymptotically stable and all motions storting in W converge to the origin as t  . Condition (ii) can be replaced by (ii)a and (ii)b as in Theorem 2.7.4.
an estimate of the region of asymptotic stability as follows: Choose a V(x) for all x and let V(x) be < 0 near the origin. Let V be the lowest value of V(x) on the surface V(x)  0. min
28
V(x)
is contained
in the region of asymptotic stability and therefore provides Note t.hat the region obtained by the and different
choices of V(x) may yield different estimates of the region of asymptotic Then the union of these regions is contained in the exact Thus, improved estimates can be obtained region of asymptotic stability.
+ x
l
X
2X2
4 X
"
l 3x2 :> 0.
Choose
V(x)
* 2x2  2xxx2 *
It is easily verified by applying Sylvester's theorem that V(x) V(x)  6 (x2  x^ 4 1 x^x2 4 x2) < 0 but in general 2.3. V(x) < 0.
We proceed to 
find the region of asymptotic stability by examining the surface V(x) It is sketched in Pig. for x^ = 1. Computations reveal the curves V(x)
c > 0
4 ;
o e&^iWo
x2 = x Then x^ = x2 3
28
The above equations are known to have a stable limit cycle. The equilibrium
point (origin) is an unstable point. If we put t  t, then the origin will become a stable equilibrium point with an exact region of asymptotic stability given by the limit cycle. we shall yet an estimate of this region. By using Theorem 2.7.5, Putting t = t
II
*1
= X
"xi
+ f(
 "
4
:
Choose
32
2 x V(x> =
, + x2
2
(yj
1J
V(x) < 0 for x2 < 3 dx V(x> * 0 on the x^ axis. On the entire x1 axis = + . and hence the x^ axis for the system. for x2 x.^ " t ^' 0 does not constitute a solution of Since V{x) < 0
Hence the only solution is the origin. '''his is easily found to be 3/2.
< /'3 the value of Vmin is evaluated by compulinq V(x) at x^  0, Hence an estimate of the
+
< 3/2
~ X2

x2 = 2b x^  ax2
3x2
We will examine the stability of the origin and try to construct a region of asymptotic stability. Take V(x) as
V(x) V(x)
 0
Examining "
ax
 3x
 ^
32
it is
outside the origin and zero at the point at the point 4b3 (2b/3,0), we
(2b/3,G)
Hence, the equilibrium points are the origin and the point Evaluating V(x)
stability of the origin. From Lhe preceding three examples, we see that the region of attraction V(x) < C can be computed either by evaluating C on the carve V(x)  0 and taking minimum value or evaluating V(x) at the nearest equilibrium point other than the origin.
AM
itturns out,
We have so far only stated the theorems with several illustrative examples. The proofs have been omitted since they involve However, a geometric
dt
" k=i
3x
Consider the surface V(x) = C, C > 0 and some point on the D V surface. The quantities are the direction derivatives cxk along the different axis. The normal to the curve V(x) is
av
qiven by the vector VV whose components are ? . represents angle of the normal to x. axis, then
A
If nx, dxk
W^ = oX k
The positive direction of the normal is in the direction of increasing V{x). Now
dt
n dx
k=1
3xk
dt
I I
k=1
dt
. represents the sum of the projections dV of the velocity vector components on the normal VV. If ^ < 0, The term T. k=1 it means trajectories move towards surfaces with smaller value of C. This in turn implies that the trajectories intersect the families of surfaces inward since surfaces for small values of C are contained in surfaces with greater values of If dv/dt > 0, then the trajectories cross v(x) = C surfaces n dxk from inside. If 51 ^r cos r;xi = r the trajectory lies on dt k k1 the surface V(x)= C. The preceding observations can be C. verified from Pig. 2.4 for a twodimensional case.
32
2.9
In this section, we will consider the stability investigation of linear time invariant (L.T.I.) Consider the system x = A x with origin as the only equilibrium point whose stability is to be investigated. plane or not. The stability of Alternatively, (2.7) can be examined either by computing the eigenvalues of A to see if any of them lie in the right half the Routh Hurwitz conditions can be = 0. However, applied to the characteristic equation given by det[A  All which arc stable. systems by Lyapunov's method.
(2.7)
both these methods do not readily give an insight into the class of A matrices This is possible through the construction of a Lyapunov function of a quadratic form which also yields the necessary and sufficient conditions regarding stability of the A matrix. The following theorem regarding the solution of a matrix equation is of fundamental importance. Consider the matrix equation ATP + p A = Q where A is the given n*n matrix, P and Q are symmetric n*n matrices. a set of Theorem:
n
(2.8)
If Q is given, the above matrix equation results in linear equations for the elements of P. If ,,...,3L are the eigenvalues of the matrix A, then the
tor all
i, j  1,2, ...,n
This implies that for a unique solution to exist, A should have no zer0 eigenvalues and no real eigenvalues which are of opposite sign. of this theorem is contained in. Hahn [33 2.9.1 Lyapunov function _for_LT.I. Systems T We choose V(x) = x P x where P is positive definite, as the Lyapunov function for (2.7). Then The proof
i+ 35(a)
V(X)
X X
x T (A P + P A)x T x Q x Since P is positive definite, by Theorem 2.7.3 the origin is asymptotically stable if Q > 0. is also a.s.i.l. theorem. 2.9.2 Stability Theorem for L.T.I. Systems Then for the Since origin is the only equilibrium point, the origin This leads us to a precise statement of the following
Suppose that Q is a positive definite symmetric matrix. system (2.7), A is a stable matrix, i.e.
negative real parts if the solution for the real symmetric matrix P of the matrix equation A P + P A = Q is positive definite. T Furthermore, V(x) = x P x constitutes a Lyapunov function for the system (2.7), having a negative definite time derivative V(x) Equation (2.9) = xTQ x is commonly known as Lyapunov1s matrix equation. (2.9). Suppose we pick Q Suppose (2.9)
We must clearly understand the implication of to be any p.d. matrix say Q = I, than one solution, that (2.9)
then if the equation has no solution or more then again the origin if the unique solution
The preceding theorem and discussion has not commented on what happens if A has eigenvalues on the imaginary axis. is stable but not asymptotically stable. The question is partly answered by the following In terms of Lyapunov's stability the origin theorem if there are distinct eigenvalues on the imaginary axis,
theorem due to B. D. O. Anderson 17]. Theorem: If the matrix A has no eigenvalues with positive real parts and
has some eigenvalues with zero real parts which are distinct, then for a given
i+ 36(a)
Q > 0,
there exists a P > 0 satisfying the Lyapunov matrix equation = xTP x and V(x)
(2.9). 0.
= xTQ x <
One of the main impediments to the application of Lyapunov's method to physical systems is the lack of formal procedures to construct the Lyapunov function for the differential equations describing the given physical system. We have seen in the previous section that for a linear time invariant system, systematic procedures exist via the Lyapunov matrix equation. However, the bulk of practical problems fall under the nonlinear category for which systematic procedures are an exception rather than the rule. to a certain class of problems. in a paper by Gurel and l.apidus as 1) Methods based on first integrals 2) Methods based on quadratic forms 3) Methods based on solution to partial differential equation 4) Methods based on quadratic plus integral of nonlinearity type Lyapunov function 5) Miscellaneous methods 2.11 Methods Based on l'irst Integrals 19) A number of investigators have worked on evolving some kind of general methods applicable A good categorixation and historical is contained 18]. We can broadly categorize the methods summary of various methods of constructing Lyapunov functions
This method of constructing Lyapunov functions is based on a linear combination of first integrals of the system equations. In turn, this is based on Lyapunov's original idea that total energy, in the case of a conservative system, could serve to define the stability of an equilibrium point. We shall first define what we mean by a first intcqral. Consi der
x
= f
i< x r x2  V (2.10) = f (x) , f (0)  0 (or simply integral) we understand a ditterentiable defined in domain D of the state space such that assumes a constant
i.e.
By a first integral
system. A necessary and sufficient condition for (2.10) to hove a first integral is given by the condition n
3f.
II
37
*<'*
1>
il l second order systems. Example: Consider
x
(2 1
for
= X
2
(2.12)
x2 = 4xJ  4x
It is easily verified that condition (2.11) is satisfied for this equation. Now (2.12)
d
can be manipulated as
X
*l 2
dx
4 x 3 _ ^ Multiplying and
integrating we get
x2
G(x1,x2) 
+ /q
x 4 (x1xJ)dx1
Since G(x^,x2) it constitutes
as the first integral or energy intcqral of the system.  0, and G(x^,x2) is positive definite around the origin, a Lyapunov function. The origin is stable.
The method of first integrals is limited to systems which are conservative and which in addition satisfy condition (2.11). 2.12 Method of Quadratic forms (Krasovskii1s Method) [10]
In this type of method, the Lyapunov function is of the form T T x or * h L P()* Consider the autonomous system x = f (x) , 0 f (0) (2.13 )
Assume that f(x) has continuous first partial derivatives and define the Jacobian matrix
3f
3f n
3f
ate"'  ^
(2.14 ) Df n
Tlx
38
Krasovskii's Criterion If there exists a constant positive definite matrix P such that the matrix 0(x) defined by
2<x) * g J(^)
2' txP
is negative definite, then the origin of (2,13) is asymptotically stable in the large. Consider the Lyapunov function V(x) = f P f
which is positive definite in the f space. Since there is a one to one mapping between xspace and fspace, V(x) is also positive definite in the xspace. The derivative of V(x) is given by V(x) rule f(x) = J (x)x  J(x)f (x) Hence = fTP f + TP f By chain
V(x) is negative definite since the term inside the brackets is negative definite. Example:
x
Hence,
~axl *
2 a > 1 (2.15 )
x1
x2
x2
The condition for a.s.i.l. of the origin is that Q(x) be negative definite i.e., 2a < 0 and 4a + 12ax2  4 > 0. are satisfied. Since a > 1, both these inequalities Hence, the origin is a.s.i.l.
39
2.13
[11]
function is that both V and V may be determined from it. 3V . dt *1 <VV, x> V = /" <VV, dx> 3xx
n n
and
The upper limit x in the integral, indicates that the line integral an arbitrary point in the x space and is independent of the path of integration. x The explicit form is
v=/
is to
oi
vv
i (y i'"" 0 >^i *
s 2
x q 2(x1,Y2 ,o...O)dr2
x dY
4 ....
/0n
(2.16)
.. nl'V n
where VV^ is the ith component of the vector VV. It is shown in standard texts on vector calculus that for a scalar function V to be obtained uniquely from a line integral of a vector function y v , the following
n
3VV
TST*=?
These are also necessary and sufficient conditions so that the scalar function V be independent of the path of integration. In the threedimensional case, Eqs. (2.17) are identical to those obtained by setting the curl of the vector VV equal to zero. The procedure of constructing V(x) is to determine a gradient VV such that
i) The n dimensional curl of VV is zero, ii) V and V is determined from VV such that V > 0 and V < 0. If V < 0 then we must ensure that it If such a VV is not zero along any solution other than the origin. consists in assuming 9V of the form
II REVIEW Of STABILITY
THEORY
40
a12(x)
W=
lnl*> (2.18 )
n2 W
nn ...
plus a " J K variable term o^jy* For V to be positive definite in the neighborhood of the origin, the diagonal terms (x) arc assumed to be of the form au(x)  ii(xi) = a..k + iiv(xi)
>
(2.19 ) In
with
this manner after integration, the rise to terms such as a... ilk =2 and
where u^ is a dummy variable of integration. Further knowledge of the unknown coefficients in TV is obtainable from an examination of the generalized curl equations 3a. . (x)xx + (a . ., x  + a  . (x) x .) ijk ] rjv v  J
y
5xj
Ox. J
dX
dX
n
3
la 4 U x
3VV,
"9xT
ax
1 (x)x1 55" + .
jikXi
3 a 
From
(2.17),
(2.20)
are equal.
Since
it follows
ijk
The other terms in' VV are determined jointly from the curl equations and dV/dt. Since dV/dt < 0, (2.17) an attempt is made to make it negative
II REVIEW Of STABILITY
THEORY
41
semidefinite in as simple a way as possible. if indefinite terms in dV/dt are eliminated. assume 1 M where k is initially assumed a constant. 1 .
the above equation, the remaining terms in dV/dt must be forced to cancel This is accomplished by grouping terms of similar state variables The a^j's
are
assumed constants
unless cancellation or the generalized curl equations require a more compliGrouping of terms is governed by the restrictions For example, on the u.'s as stated above. A J if in a third 2 3 ft x and x x order system dV/dt contains the terms a^x^xj* i2 2 ~ l 2' the indefinite term "X^Xj could not be grouped with a2_ i x i x2 since a,, can only be a function of x,However, if 3 2 cou x x were rou x ~l2 y Ped with a^2 2' ^^ ke eliminated by letting
= x x
j 2"
T e c
to
force cancella
tion is not arbitrary because the generalized curl equations (2.17) must also be satisfied. variable, variables. If dV/dt cannot be made semidefinite in one state then one may try to make it semidefinite in two or three Another possibility is to make dV/dt neqative semidefinite the region of asymptotic stability by filling
in as large a region as possible around the origin in which case, we can try to find an estimate of the largest possible V curve using Theorem 2.7.5. A formal fivestep procedure for the application of the variable gradient method is given below: I) Assume a gradient of the form (2.1fl). The
a i^'s
may be
written as though they are constants until the need arises to allow them to be more complicated. 2} From the variable qradient, form dV/dt = <VV, (x)>. 3) in conjunction with and subject to the requirements of the generalized curl equations semidefinite. (2.17), constrain dV/dt to be at least negative
42
4) From the known gradient, determine V. 5) Use the necessary theorem to establish stability, asymptotic stability in the large, or region of a.s.i.l. In the above procedure, the most difficult step is step 3. 3 can be conceptually broken down into smaller steps as 3 (a) Constrain dV/dt to be negative semidefinite. some contraint on the coefficients. (b) Use the curl equations to determine the remaining unknown coefficients. (c) Recheck V because the addition of terms required as a result of 3(b) may alter V. Example: Consider the system of equations x, 3x*  g(x.)x.
&2 = ~2xj
In fact,
step
' gtx^lx
1:
The procedure outlined above will be carried out step by step. Step The gradient function is assumed to be + a a VV * + a
22X2
11
 129'xl* ~
2a
21
+ a,,g(x.)]
 x*{2a Step 3: The simplest manner in which dV/dt may be constrained to be negative semidefinite is it dV dt " ~allgtxl)xl " 2
2ct
" 2i ~ ^. 2
+ X X (3a
Then
22X2
l2
ll
+ a
22
22
43
ll
"
g(X }
l Thus,
the gradient
is known to be 22
VV =
ft
22*2
a ~T~ y^l^l ^!
1 +
f
.
Q
ft
22T2dV:
With fractious
) T 1dy1
as [x^ Lhcn V(x) is not only positive C definite, but also V(x) * represents closed surfaces for all C in the whole state space dV/dt is negative definite in the entire state space. Step 5: On the basis of Theorem 2.7.3 on a.s.i.l., the system is globally asymptotically stable.
II
In these steps, particularly in step 3, no apparent use was made of the generalized curl equation.
was a
However, since VV
they were implicitly used, since does not contain x and VV docs
constant equal to zero and on the basis of the curl equations i&gj. was
3
v
2.14
This method enables us not only to generate a Lyapunov function, but also construct a region of attraction or an approximation to it. and when the solution to this p.d.e. The crux of (p.d.e.) When the method lies i n solving a linear partial differential equation
obtain a unique Lyapunov function and an exact stability rcqion. the solution and we get an approximation to the exact stability region. this method is amenable to computer solution. theorem. Theorem:
Let U be a set containing the origin.
i s not possible in closed form, a series solution is possible In particular, We first state Zubov's main
Necessary and
sufficient conditions for U to be exact domain of attraction is that two functions V(x) a) V(x) b) 8(x) c) V(x) and 6(x) exist with the following properti as: 0(x) is defined
and continuous in the entire state space. is positive definite in U with V f O ) = 0. Also,
the inequality 0 < V(x) < 1 holds i n U. d ) On the boundary of U, V(x) = 1. e) The following partial differentia] equation is satisfied
(2.23)
state space, we then have global asymptotic stability and the condition on is Lhen V (x) > 1 as  x  *
define another positive definite function <>(x) <t(x>  Q(x) (l+f 2) The p.d.e. then becomes n I ) 1=1
2. In another formulation of Zubov's method, the right hand side of the p.d.e. (2.23) appears as
H2)1^2
the
(2.25)
3. If we define a change of variable V =  n(lV], p.d.e. (2.24) " i~l and the becomes ^ f i (x) = <fr(x) is V. The region of
Lyapunov
function
~xl
+ x
+ x
i^xi+x2^ 2
2 Substituting in (2.24), 
+ 
(xrx2+x^x2x2) = 2(x2+x2)(lV(x))
2 2 It turns out that V(x)  x. + x.. satisfies the p.d.e. Hence ~ 2 2 the required Lyapunov function is V(x) = x^ + x2 and the stability boundary is given by xl * x2 ^
=
~X2
Dxx
(xj 3x2
~2
{ ( i  v j n + x2 * (2x^x2x1)2]1/2:> 2 j
Let
Clearly 0\{xyxg) satisfies the condition of the theorem. The solution to the p.d.e. is given by 2 "X2 V = 1 + exp(5 2 "Xl 2(lxax2) V (xlfx9) vanishes at the origin and it is equal to 1 on the
x x Hcnce
curve j 2 ~
In qeneral, we cannot expect a closed form solution to the partial differential equation in Zubov's method. expressible in an analytic manner, e.g. If the nonlinearity is power
4 8
series, then Zubov's method can be used to construct the Lyapunov function as well as approximate the region of attraction. Let x = f(x) be expanded as (2.27)
contains terms of second degree and higher and A is the linear We choose <Hx) to be a positive definite quadratic form. The
3V * f^tx)
(2.28)
v x 2< )
+ v 3<x)

is quadratic in x and
v x m< )(m=3,4,...)
i.e. >j) = ^"Vfx) for any constant To find V (x)(m2.3....) we substitute the system m (2.27) and V(x)
v m( x
given by
^2*^
(2.29)
in (2.28).
)*
(2.30)
ta
av
i^r
.<s> V*>
(2
31)
m n 7)x '
[A )
x]
fx)
which simplifies to
m=3,4,... (2.32
>:
v m_j
(x)
depends upon the particular problem under consideration. converge to the exact stability boundary uniformly. a positive definite quadratic function.
in (2.28) by means of a positive semidefinite quadratic function instead of arc still guaranteed to get a positive definite V2(x). 2.14.2 Computation of Region of Attraction
The procedure is outlined below with the computer implementation in mind for computing the region of asymptotic stability.
4 8
* $
2 W ^or
xe
^2'
entirely inside the region of attraction of the equilibrium point and the curve defined by V^(x) B lies entirely outside the region of attractionThese two assertions imply that the boundary of the region of asymptotic stability lies in the region defined by a < V2(x) < B , V2 + + V^,..,,etc.
We can extend the above reasoning to V2 + To generalize, define V(n)(x) = V2(x) Let W^n* (x) + V3(x) + ... Vn(x)
Specifically, we ensure that W*n'(x) comprises all points of zero v'n^ (x) which define boundaries between regions of positive and negative V*n*(x). value of V W
(n) (n)
Let
be the smallest on
n)
on W
{n}
(n)
and
Let A
say that the curve V*n^ (x) = u*n* is wholly contained in the region of attraction and the curve v'ft^ (x) attraction. successively compute a^n' and the region A^n'. One would be 3 * n ^ wholly lies outside the region of
In implementing the procedure for a numerical problem, we wo (n) tempted to infer that as n takes on large values, A approximates better and better to the exact stability region. However, this is not necessarily true/ i.e.
2.14.3
x +
p (m
m.,
m,
*" (2.33)
% * mx + m2
... itl
(i=l,2f . . . ,n)
49
where the a^ and pi(n1 ,m2, . . . ,mn) assume ii of the form homoqeneous of degree i. n 3V7 n + + ... Then for
Also
is a term
v we 2(x)
(2.32) become
m=3'4( 2
'35)
is formed from and the already known V.'s. m ' l (2.34) and (2.35),
If we compare like terms on both sides in for coefficients of Vm(m2,3,...). truncated up to a certain value of m. 2.14.3 Computation Example [12].
then a set of linear algebraic equations result which may be solved The series is generally
This is obtained from the usual Vanderpol's equation by putting t = t so that we have a region of attraction around the origin. Introduce the variables defined by
II
RKVIKW OF
STABILITY THEORY
51
3.0 2.0 n 6 n  10 14
NOl fc : VA N DER POL E Q UA TI O N f. * 0.7 LIFNARD PLANE SOLUTION J ____ . __ I ______ I ____________ 2.0 Pig. 2.5. 1.0 0.0 X 1.0
Stability boundary H P given by approximate solutions t o Zubov equation. (Reproduced from Kef. [12].
 X
(2. 37) x,
+ X
(x 
nd thus obtain
x.
=
2 " '
X
(x
l " ~: (2,38)
(2.38), wc have the state traicctories which This contrasts with the usual The results can be transferred to
plot x dyainst its (negative) time integral. procedure of x versus its time derivative. the usual phase plane by solving for x
from (2.37).
51
"
I d j*2 k0
x
JX 1 2
x K
(2.30)
=
+
(d
(d
= x2 + x2.
now
(_Xi)
.v(3)
(2.41)
2
x tx
9V
2 ' i 2
(x +x )
2 (2.42)
W < 2sv
i>
385 "
( Xl>
av
3x7
(x _cx > +
m (xi> = 0
The appearance of 0 term on the right side of the second equation in term in the nonlinearity and, furthermore, . In Rcf. region of attraction computed is Laken as 0.7. of radius /3. For n=6, (Fig. 2.5)
(2.42)
for terms of degree 3 is not surprising because there is no second degree i> ( x ) has been chosen as <K ( x ) ~ x and as discussed in Sec. 2.14.2. c improved, but 112j, computations have been done by taking tf2(x)
For n=2, the stability region is found to be a circle the stability boundary is not On the oLher hand, n=10 approximation
contained in the curve for n2. encloses both the n=2 and n=6
approximation and is in turn completely enclosed by the n = 14 approximation. Thus, a surprisingly high order approximation must be used in order to improve upon the quadratic results. References T12] and [121 contain a good discussion regarding the numerical aspects of the solution. 2.15 Absolute Stability  Popov's Method
52
yo far, we have been considering general nonlinear systems without any restrictions on the nonlinearities. We also found that for such systems there is no systematic method to construct Lyapunov functions in order to ascertain stability. In addition to the classical method based on first integrals of motion and Krasavoskii's method, two other popular methods were discussed, namely, variable gradient and Zubov's method. However, if the nonlinearity is such that it lies in the first and third quadrant or in a sector thereof, then a systematic procedure to construct the Lyapunov function is possible. The problem therefore is to investigate the In 1944, Lure [141 constructed a Lyapunov stability of such systems.
function for such systems consisting of the sum of a quadratic form in the state variables and an integral of the nonlinearity and derived therefrom a set of nonlinear equations called Lure's resolving equations. conditions for global asymptotic stability of the nonlinear system. significant breakthrough in this area of investigation came from the contribution of V. M Popov in 1962 [2]. Popov outlined a frequency domain Kalman [15J and criterion tor the stability of a nonlinear sysLem containing a single nonlinearity lying in the first and third quadrant. Yakubovitch [16] established the connection between Popov's frequency domain criterion and the Lyapunov function of the type of quadratic form plus integral of the nonlinearity by proving that satisfaction of the former is a necessary and sufficient condition for the existence of the latter. The A existence of a set of solutions tor such equations constituted sufficient
Subsequently,
nonlinear systems with multiple nonlinearities as well discrete time systems, etc. engineering literature. 2.15.1 Single Nonlinearity Case Consider the [173.
where A is a n*n stable matrix, x, b, c arc nvectors. can be oast in the block diagram as in Fig. verified from (2.43)
$ () is a (2.43)
nonlinearity which l i e s in the first and third quadrant. The system transfer function with a nonlinearity in the feedback path.
53
I f A is a stable matrix, i.e. all eigenvalues are in the left h a l f plane, then G(s) has a l l poles with negative real parts.
WH
Special cases o f interest arc when G(s) has poles on the imaginary axis. will single pole at the origin. then i s of the form x
i
consider only the simplest of these cases, namely, when G(s has a In terms of the state space description, t h i s The state space description
A 0
b 1 (2.4 5)
0=
*<o)
o c x + d In addition to A being a Hurwitz matrix in both (2.43) and (2.45), we will further assume that ( i ) t h e p a i r (A,b) is
Fig. 2.7. Block diagram for Che system controllable, and (ii) {A,c ) is observable. The nonlinearity satisfies the sector condition, 0 < <Mu) i.e. < ko? (2.47 ) for the system (2.43) and
55
0 < ^ (o)
+ Re{ ( M >>q)G(
} > 0
for all
w > 0.
(2.49)
In the case of the system (2.45), an additional necessary condition, namely, d > 0 is to be satisfied. It is noted from the transfer function that This condition is physically equivalent (2.45) is stable if $(a) = a(e > 0 and d is the residue of G(s) small). 2.15.3 Geometric Interpretation at s = 0.
For the sinqlc nonlinearity case, a simple geometric interpretation can be given to the Popov criterion. G(jw) Then (2.49) ^ X(fii) + j Y((o) becomes + j Y(<n)] > 0 > 0 (2.50) Let
^ + X(w)
 qu> Y(oi)
 X <J.I)
and
Y*(OJ)
= <o Y (a>)
IT REVIEW OF STABILITY
THEORY
57
*
Then we can define modified frequency response G G*(w> Expression = X*(a>) + j Y*(<it) (jui) as
(2.50) \ 0 X*(w>
(2.51)
frequency response plane as impiyinq that there must exist a line called the Popov line passinq throuqh the point ( 7 , 0 ) K 1 * having a slope so that G (jui) lies to the right of this line (Fig. 2.3). Thus, absolute stability, i.e., a.s.i.l. of the (2.48) can be ascertained in a simple manner An origin tor nonlincarities satisfying the sector condition (2.47) or analoqous to the Nyquist's criterion for linear systems. additional benefit of Popov's criterion is that if we can establish absolute stability, then we also have a systematic procedure to construct a Lyapunov function. type. equations sector case, in Refs. This Lyapunov function is of the quadratic plus integral of the nonlinearity Systematic construction of the Lyapunov function is (2.52) and i.e. (2.53) which are written for the infinite These equations form part of the possible through the solution of a set of nonlinear algebraic k = >.
(2.52 )
1 n [fl(o b+d>]
T
(2.53)
Multivariafcle
nonlinear
where
q = gjfeg ,
(n1)
vector.
not the special case of the latter. If Popov's criterion is satisfied, we know q. of a, 6 and u, a p.d. solution and the Lyapunov function is then of the form For the system (2.43) V(x) For._the system V(x) In arriving at = XTP x + q (2.45) = xTP x (2.54) t ^ dr. 2 + q f *(o)da and (2.55) the choice of (2.55) <Mo)dc (2.54) Then by an arbitrary choice (2.52) or (2.53)
 1 and cc =
has
The solution of nonlinear algebraic equations can be circumvented for the single nonlinearity case by what is known as the Kalman's construction procedure [15]. The various steps in the procedure are: 1. Find q from the Popov Criterion. 2. Write down the following polynomial of order 2(n1) in in.
Vim) 
(ji.i)tp (>') 1
3. Factorize W(w) * 0 (jw> O(ju) n for the system (2.43) and (r,l) for sysLem coefficient /r where r = q cTb for 5. Define v(z) (2n2) (2.43) = 8(z) and q(d+cTb) + /r ifMz) for and arrange this
polynomial of formal order 2n for the system (2.43) for the system (2.45) in ascending powers of z. or
6 Define a real nvecLor u for the system (2.43) coefficients of the above v(z). matrix equation ATP + P A  u uT
(n1)vector for the system (2.45) with its components being the Solve for P using the Lyapunov
(2.59)
The Lyapunov function is then given by either 2.16 Multivariable Popov Criterion
(2.54) or
(2.55).
There are several versions extending Popov's criterion to systems containing multiple nonlinearities. Perhaps the most popular among these is using the MooreAnderson Theorem 118]. Consider the system in Fig. 2.9 where W(s) is an mm matrix transfer
function such that W(<*>) = 0. assumed to satisfy the properties i) ii) (ii) fk(a> = fk(ok) fk.<ok).ok > 0
k=l,2, . . .m
implies that each nonlinearity lies in the first, and third quadrant.
Let A, B, C constitute the minimal realization of W(s), i.e. W(s)  CMsIA)"1! The state space description of the system is given in the form x A x  B f(c) (2.60) a = C x where A, B, C are nxn, n*m and m*n matrices respectively. If there exists real matrices N and Q such that (2.61) is asymptotically stable in the
(N + Q s)W(s)
= Z(s*) for Re s > 0. iii) Z (s ) + Z(s) is positive semidefinite for Re (s) > 0. (i) and {ii) hold for most physical systems and condition
Conditions (iii)
(2.52)
and
(2.53)
nonlinearity case, we have a method of constructing the Lyapunov function if the MooreAnderson theorem is satisfi ed.
Theorem I181 Given N and 0 satisfying the multivariable Popov criterion (2.61), then a Lyapunov function of the "quadratic plus integral of nonlinearity" exists of the form V(x) = xAP x + fQ f/(c) Q do (2.63)
and P is obtained as a positive definite matrix satisfying the following set of nonlinear algebraic equations
m m
W W
where L and W arc auxiliary matrices. Comments i) Unlike in the single nonlinearity case, an easy geometric interpretation of the multivariable Popov criterion is not possible, ii) Extensions of the above basic theorem to finite sector [171. case are available in the literature
62
iii)
systematic solution procedure like Ralman's construction procedure Tor the single nonlinearity cases unless we resort to techniques like spectral factorization, etc. (2.64) Fortunately, is we shall see in Chapter IV, for the power system case the mathematical model results in a solution of for P easily, iv) Although not explicitly stated, the condition on A for the theorem to hold is that it be stable. The poles on the imaginary axis if they exist must be simple. Conclusion In this chapter, we have discussed various methods of constructing the Lyapunov function in a systematic manner. We have studied Lhe basic Popov's criterion, stability definition in the sense of Lyapunov and theorems which give sufficient conditions for stability of various types. discussed. both for single nonlinearity as well as multinonlincar cases, have been
II
63
References 1. A. M. Lyapunov, "Problems General de la Stabilite du Movement", Reprinted in Annals of Mathematical Studies No. 17, Princeton University Press, Princeton, N.J., 1949 (Russian Edition 1892). V. M. Popov, "Absolute Stability of Nonlinear Systems of Automatic Control", Automation and Remote Control, Vol. 22, 1962, pp. 857875. W. Hahn, "Theory and Application of Liapunov's Direct Method", (Book) Prentice Hall, Englewood Cliffs, N.J., 1963. J. P. LaSalle and S. Lefschetx, "Stability by Liapunov's Direct Method with Applications", (Hook) Academic Press, New York, 1961. J. L. Willems, Nelson & Sons, "SLability Theory of Dynamical Systems", (Book) Thomas U.K., 1970. (Book) Prentice Hall,
2. 3. 4. 5. 6. 7.
B. D. O. Anderson and S. Vongpanitlerd, "Network Analysis and Synthesis  A Modern Systems Theory Approach", (Book) Prentice Hall, Englewood Cliffs* N.J. O. Gurel and L. Lapidus, "A Cuidc to the Generation of Lyapunov Functions", Industrial and Engineering Chemistry, March 1969, pp. 3041. N. G. Chetaev, "Stability of Motion", (Russian Edition, 19461950). (Book) Pcrqamon Press, 1961
8.
9. 10.
K. N. Krasovskii, "Stability of Motion", (Hook) Stanford University Press, Stanford, California, 1963 (Russian Edition 1959). D. G. Schultz and J. E. Gibson, "The Variable Gradient Method for Generating Liapunov Functions", AIEE Trans. Part II, Appl. & Industry, Vol. 81, 1962, pp. 203210. S. G. Margolis and w. G. Voqt, "Control Engineering Applications of V. I. Zubov's Construction Procedure for Lyapunov Functions", IEEE Trans. Automatic Control, Vol. AC8, No. 2, April 1963, pp. 104113. Discussion on Rcf. 12 by F. Fallside and M. R. Patel and Reply, IEEE Trans. Automatic Control, Vol. AC10, No. 2, April 1965, pp. 220222. A. I. Lure and V. N. Postnlkov, "On the Theory of Stability of Control Systems", Prikl. Mat. i. Mehk, Vol. 8, No. 3, 1944. R. E. Kalman, "Lyapunov Functions for the Problem of Lure in Automatic Control", Proc. Nat. Acad. Sci., Vol. 49, February 1963, pp. 201205.
11.
12.
64
10.
v. A. Yakubovitch, "The Solution of Certain Matrix Inequalities in Automatic Control", Dokl. Akad. Nauk. S.S.S.R., Vol. 143, 1962, pp. 13041307. K. 8. Narendra and J. H. Taylor, "Frequency Domain Criteria for Absolute Stability", (Hook) Academic Press, New York, 1973. J. B. Moore and B. D. 0. Anderson, "A Generalization of the Popov Criterion", Hour. Franklin Institute, Vol. 285, No. 6, June 1968, pp. 488492. M. A. Aizerman and F. R. Gantmachcr, "Absolute Stability of Regulator Systems", (Book) Hoi denDay, Inc., San Francisco, U.S.A., 1964. (Revision Edition 1963)
17. 18.
19.
CHAPTER III
3.1
Introduction
As a preliminary step Lo the application of Lyapunov's method for power systems, wc will develop the necessary mathematical model of a multimachine power system in this chapter. The bulk of literature in Lyapunov The mathematical model can be stability analysis of power systems rests on a mathematical model involving a number of simplifying assumptions. retained in the form of a set of second order nonlinear differ ential equations or conveniently be cast in the Lur6 form with the nonlinearities in the first and third quadrant over a region around the origin. computation of the region of attraction. Both formulations are Towards the end of the This enables the systematic construction of the Lyapunov function and also the equivalent and are prevalent in the literature.
chapter, we present the mathematical model relaxing some of th simplifying assumptions and which give rise to multiplicative nonlinearities in the mathematical model. Since synchronism is essentially a physical phenomena associated with relative motion between synchronous machines, some sort of reference angle other than the synchronous icference frame angle is necessary. two possibilities here. reference anqle There are We can take the angle of one of the machines as The other
(generally the machine having the largest inertia) and to inertia weighted
measure angles of other machines with respect to this machine. possibility is to define a center of angle proportional mechanical systems.
angles of all the machines, a concept analogous to the center of mass in All machine angles are then measured with The state space models based on both The correct dimension of the
state space for the different models is derived from physical as well as control theoretic concepts of controllability, observability, minimal realization theory, etc. The hurt? form of equations are then derived which constitute the basis for systematic contruction of Lyapunov functions and also Cor applyinq the Popov stability criterion. Doth the cases of zero and nonnegligible transfer conductances are considered as well as zero and nonzero dampinq. Finally, the mathematical model for the multimachine case including the flux decay effect is considered. 3.2 Transient Stability Problem Restated
By transient stability, we imply the stability of the system to maintain synchronism under sudden and major impacts. transmission line, disturbances. A fault on a high voltage loss of a large generating unit, are examples of large major disturbance basically creates an Consequently, the power (i.e., mechanical input power minus the
of a major disturbance.
electrical output power) differs considerably from generator to generator. Some of them will have net accelerating power while others have decelerating power. As a result, the rotor angles of the machines accelerate or This is called 1.1 and 1.2) If the rotor angles are plotted as a function decelerate beyond the synchronous speed for t > 0. "swinging" of the machines. of time, therp exist two possibilities. (See Figs.
i) The rotor angle increase together and swing in unison and ultimately settle to new angles. Since the relative rotor angles do not increase, we say the system is stable and synchronism is preserved, ii) One or more of the machine angles accelerate faster than the rest of them, so that ultimately the rotor angles of the faster group continues to increase indefinitely. The relative rotor angles diverge. Such a mode
of behavior is classified as unstable or losing synchronism. If the fault is cleared quickly, the tendency to separate may be arrested throuqh the postfault network by a redistribution of the energy gained by the system during the faulted period. The maximum time through which a fault can be allowed to remain without losing synchronism subsequently is called the critical clearinq time t cr All of the above concepts which are mainly physical have to be translated into mathematical ideas before Lyapunov's theory can be applied. this first by developing the mathematical model. 3.3 Mathematical Model We do
Under disturbed conditions, at each machine, we can write the swing equation as follows: &
(w
K.F.,i'
ai 
p mi
V
(i=1 2
 ................... n )
(3 1
>
. = the kinetic energy of the rotor in Mega K"E'1 = the damping power in MW = the mechanical power input in MW Joules
P . gi Equation
(3.1) merely states that the accelerating power for each machine Let us denote the synchronous frequency Hence,
is balanced by the increase in kinetic energy of the rotor and the power absorbed by the damping forces. by f^. frequency f^ of each machine. The kinetic energy is proportional to the square of instantaneous
" wK.f:,i
<r;>2
where W . is the kinetic energy of the i^ machine at the i\ > is i 1 synchronous frequency fQ. (f +Af.)2 K.E,i K.E,i o since the deviation of \ from fQ during a transient is small. Let respect to a synchronously rotating reference frame both in electrical radians.
6 f2
2Af. K.E,i f be
the rotor angle with respect to a fixed reference and 6^ the rotor angle with Thus i  i 
V
da
2
dS
and
dF 
Ir  "o =
UK
*<fi"fo> = 2"
if
i = i
to.
We shall make the assumption that the damping power P^... is proportional the frequency deviation .
system which is due to the asynchronous torques between the machines and is proportional to the velocity difference between machines. we can write P,. = D.u>. + (w.tt),} di x i
j
t
ij
d6= Di
n
+
MF
A Dij
<inr  at4
Although subsequent treatment can be carried out with the above representation of P.,, for ease of mathematical analysis, we will retain the terms only and assume the asynchronous torques to be aero. the expression for W
where H
Substituting
^ ^ in (3.1) .
Base^MVA
p.u. of sec.
In the literature, H and M are both referred They only differ in the units, H having units M is convenient in the
as.
.  p mi
. gi
(i=l,2f. ,n)
(3.4)
nonlinear differential equations of the electrical part of the synchronous machine and the algebraic equations of the transmission network and the synchronous machine [1]. Fortunately for our work, such a detailed We therefore make representation of the synchronous machine is not needed.
a number of simplifying assumptions as discussed in Appendix I and summarized below which yields an analytic expression for P^^ in terms of the 3j_'s. 1. The network is assumed to be in the sinusoidal steady state. compared to the electromechanical frequency of oscillation. 2. The synchronous machine is represented by a voltage source of constant magnitude determined from the steady state conditions existing prior to the fault (prefault load flow), in series with a reactance which is commonly called the direct axis transient reactance. 3. The phase angle of the voltage behind transient reactance coincides with the rotor angle 6^. This important property stems from the approximate model of the synchronous machine discussed in Appendix I. 4. Loads are represented as constant impedances based on the prefault voltage conditions obtained from a load flow. 5. The mechanical input power P^ is assumed to be constant and equal to the prefault value during the time interval of interest which is of the order of 12 sec. With these assumptions, wc now proceed to derive the relevant mathematical model. Assume the transmission network to consist of n+m buses of which the first n buses arc buses where generators are connected and at the other m buses only loads are connected. We can also have loads connected at the generator buses. n m The nodal admittance matrix This
implies that the time constants of the transmission network are negligible
(3.5)
69
Each of the generator buses is now augmented by the generator representation consisting of a voltage source 3.1). The internal = E^ in series with the (Fig. E^ and transient reactance which we shall represent by the admittance y. transmission network buses n+1,...,2n+m. generations from the load flow data network of 2n+m buses whose nodal n y Y ^BUS 0 n y m 0 *2 *4 [2,3]. Prior to the fault,
JS* are computed knowing the voltages at the generator buses and scheduled Deleting the voltage sources
Itc.
and denoting the generator admittances by the vector y, we have a transmission admittance matrix Y is given by
h+1
n n m
(3.6)
Ill MODELS
MATH
KMATI
CAT.
71
Wc next take into consideration the loads. these nodes and the yround given by
P
Li " ^ G,.i V. x
Li where
i=nt l
2n+m
(3.7)
bus obtained from pretault load flow and PT. + j Q. is the th load at the i bus. Thus, Lhe overall modified Y,lf, matrix is given as 3.2) n Y lilJS n y. m 0 *2 (Fig.
Si
0
n n m (3
*Lg *3
= Dia<
So far, we have retained the topology of the network in by eliminating all buses except the first n buses.
We now perform the matrix reduction on the fR0S matrix in (3.8) This
elimination is justified for our work since we are primarily interested in the variation of as a function of time and The y in (3.8) not the bus voltages from the stability point of view. sources are connected,
is partitioned as shown below by separating the internal buses where the voltage from the rest of the buses where no sources are connected.
71
Q 2n + m Fig. 3.1. Noda.l admittance matrix I&us of the transmission network augmented by generator representation with internal nodes 1,2, ...fn.
Oi
O n
BUS
O
Fig. 3.2.
2n + in
Modified nodal admittance matrix Y for augmented network after converting loads as constant impedances (Eu. 3.6).
Fig. 3.3.
Nodal admittance matrix Y ^ at the internal nodes after eliminating all physical "feuscs (E G . 3.11). n+m A *B n
Y = BUS n+m
(3.10 )
A
Xred
matrix a
*". *ne
ll
MATHEMATICAL MODELS
72
first n buses as
1^
W? A
WHGRG
*red
Thi8
*A " B
^V
it
(3.11)
*red
matrix is the internal bus description of the system 3.3) and although if masks the topological aspects, in Eq. (3.3) in terms of S^'s. Now we 13 (3.3)
[Fig.
have the multiport description at the internal nodes as shown in Fig. power generation is Pg. Let
Yj = G . . .
with voltage sources connected across the ports. The expression for real
RC
[ E i It]
= RelE.
Y * . Jj
i=l, 2 , ...,n .
E . 
lY.jl/.^
where
= V^ebfl^j
and
Then
P
oi "
i!2 Gii
Jx
+
Ui l KjYijIcoS(*..(6.6.)]
E.I 1
E.IE.!Y. I V
11
jjjjj
(cos^oos^fi..) Define
lE.llE.ilY.^COS^ =
+ s.iri()i^sin(6i6^) J
(3.12)
EilEjG..
5 D..
Then
V
Substituting (3.12)
i2 Gii
jA
KijSin(6ro.) * D.eosU.o.)]
in
(3.4), we get
73
d6.
\b\2 G.,
13 P
is the basic form of the socalled swiny Prior to the disturbance, At the instant of the power
fif
system is in a steady state with jS^ assuming constant values and o. = 0 for all i. fault, t = 0+, these initial conditions are valid since neither the rotor angles nor the frequency changes instantly. We have two sets of differential equations for t > 0, one for the faulted state until t t a n d the other for t > t ,. The difference cl cl between these two sets of differential equations is in the transmission configuration, i.e., this being reflected in the Y ^j parameters of the
Xre<j matrix,
it effects P ^ only.
Faulted State
d\
Hi^ +
d5
P
ilT"
mi
' "li
 Pi " Pci
<
t
6 (0>
fccl
(3"14>
"
^i
(0>
'
i TT
+ D
d P oT = mi
 Pqi
= Pf 
fc
cl'
6.(tcl)
'
(3.15)
df
determined from C3.14) i=l,2,...,n The problem of stability can be stated as follows. (t) assume constant values and 6^(t) is then stable. finding t Otherwise, the system is unstable." "Given tc^f does fi^ The problem of
We at once notice the mathematical ramifications of the above stated problem. and We have basically two sets of ri.e's (3.14) , and we are c J. (3.15) which have a discontinuity at t = t
for stability of the origin or the equilibrium (3.14) (3.14) but rather boundednews of the trajectory and (3.15). Thio ie a rather difficult
74
has to be repeated for different values of t , so that we c i finally find a t  , such that the system is stable *or t ,  fccl ' cl and unstable for t j+6 where f > 0 and small. Such a value of t , is called the critical clearing time t . Repeated simulaci cr tion to find tcr for a single fault can be expensive computationally since in a large system there are several faults to be considered. Typically, an experienced engineer does 34 repetitive
Lyapunov's method addresses itself to the question of avoiding this repetitive simulation and having just one simulation to find tcr directly. If tfir were to be estimated thus accurately, savings in computer time for planning studies. Since initial conditions for the post fault system t = tc^) correspond to the terminal values of the solution of the faulted system (3.14) (3.15) and fi. (at from it will mean a substantial
at t = fc^,
it would
seem logical to construct a region of attraction around the post fault equilibrium point and for stability, the terminal state of the faulted system (3.14) examine if lies within (3.15). system
the region of attraction of the postfault system defined by the inequality V(x) providing the points where v(x) post fault system. until V(x)
Theorem 2.7.5 provides a basis for estimating the region < C for the postfault such that inside the region v(x) > 0 and V(x) < 0 or V<x) < 0 = 0 are not solutions of the The faulted equations are then integrated . c r yields sufficient conditions, the region defined by V(x) < C is contained in the exact region of stability and hence the computation will yield in general values of t less than the actual value. be resolved. a given fault (disturbance) There are two issues to The first one is how to construct a region of stability for and the manner in which it is cleared, i.e. The second one is regarding the Computation of C, so that Since Lyapunov's method
satisfied is an estimate of t
construction of the Lyapunov function V(x) itself since different Lyapunov functions will yield different values of C. accurate estimates of t. _ o r may be obtained, has been a major bottleneck in the Initial research tended fault locations Since the mode of
practical implementation of Lyapunov's method. and ignored the faulted equation (3.14) instability is determined totally.
70
computation of C which ignored (3.14) will yield conservative values of t for certain faults and acceptable values for other faults. in Refs. 12325] 3.4 research workers for a long time until very recently when the work reported of Chapter I removed this conservativeness.
will pertain cither to the faulted state or the postfault state depending respectively on whether the equations are to be integrated or a Lyapunov function has to be devloped. In the literature, there are several types of models proposed and often (3.13) as it is in which case
there is difficulty in reconciling results obtained from different models. One school of thought prefers to treat Eq. the postfault stable equilibrium point (SEP) is at a point other than the
origin in the state space and the Lyapunov function is not zero at the origin. The other type of model transfers the postfault SEP to the origin by a coordinate transformation and V(x) is zero at the origin. Moreover, the We equations in this case can be put in the standard Lure form well known in modern stability theory and control theoretic approaches are possible. follow the latter approach. angle is always required. machine angle as a reference between them. 3.4.1 Define State Space Model Using Machine Angle as Reference In all power system problems, a reference We can have the state space model based on (MAR) or center of angle as a reference (COA).
We discuss both types of models in this chapter and point out the connection
*
x Then,
...... V
ex
w.
+ P.
1
1=1,2,...,n (3.16)
This mode has 2n state variables with rotor angles 6 . deviations works in Rcfs. as state variables. [13] and [15]
functions for multimachine power systems using MooreAnderson Theorem and Kalman's construction procedure respectively. functions are valid i f we invoke the partial stability concept
76
functins * angular differences only and, hence, we must take angular differences of machines with respect to a reference machine as the angle subvector while the velocity subvector remains unchanged. 110,16], Chap. I) Thus, the proper description of (3.13) (2nl) dimensional state vector 2
=
'Vw2%!6rWSn5nlJT
where the n**1 machine has been taken as the reference machine. This results in the state space model i ui. = 7T tu. l i *i "
D
I
+ P.
P ei
' (3.17)
"i ~
i=1 2
' . n1
PL
=
A
for
i
all i) now be x = [u,a) *wui 1 n 2 ,u> 16,6 r6'fi ....6 fl J n' n1 n 1 n' 2 n n1 nJ T or zero damping case (D^ = 0 for all i). The state variables will
The state equations will be i i=l,2,...,nl  (fi.*n) = ((v,n) i=l>2 ...... l Thus, both for uniform and zero damping there n (3.18)
case, the order of the state space model is 2n2. In the literature,
has been considerable discussion regarding the correct order of the state space model for cases of nonuniform damping, uniform damping and zero damping 16]. While in the preceding discussion, we have resorted to purely i) controllability and observability notion This will be explained physical arguments, the derivation of the correct order of the state model has also been done via 15], and ii) minimal realization theory [7].
in the context of Lure formulation in section 3,6, 35 State Space Model in Center of Angle Reference Frame
An alternate way of viewing the model is through what is called the center of inertia or center of angle (COA) formulation. This was proposed by 110]. As we shall Tavora and Smith 18], Stanton [9] and Fouad and Lugtu function approach.
see in Chapter VI, this formulation is particularly suited for the energy The formulation is analogous to the center of mass
77
Consider the swing equations (3.13) notation 6 = and 6 "fir at dt in (3.13), we get n Z . ) 1=1
1
M.
.. 6.
n , %
n D. 6. 
n Z
P.
P (3.19
il
i=l
i=l
ei
n
H = M. i=l
1
Also define Lhe new rotor angle %, with respectto 9.  S.  fi.... Let 6. = to. = w.  u> where u; that i i o _ x i o o
1
as = 6 i
Note
not all 0.*s and o>^ s are linearly independent since from the definition of it follows that n E ) 1=1 n M.e.
1 1
 0
and i=l
I
1 1
M.O.
= 0 (3.21
Vo " "Vo
 j "Vi Jx
+
i
2 E Z D . .cos6. . (3.22)
n 11 13
n i1 n PC0A
I
1
D. n ^. % n "
Define )
Then
..^
D..coSS.. (3.23
COA
(3 24)
The dynamics of the center of angle are governed by equations in the variables 6^ now become manipulation)
78
M.
+ P
(3 25)
"
i "
ci "
COA
....
79
Equations (3.24) and (3.25) therefore represent a set of coupled second order differential equations.
(n+1)
At the end of the transient period, if the system settles to a new steady state, we should have Q 0> frequency
9^ = 0, 8. = 0. where 6
J
is w = to
Y
+ 6
.*
is the
O
Vo
" PCOA
(3 26)
"
and is the inertial angular frequency. As we shall see in Sec. 3.6, at the postfault SEP, P_ = 0 and, hence, 6 ^ original 3.5.1 frequency. Simplifications of CPA Equations for Uniform and Zero Damping Case = 0 so that the system regains the
The COA formulation yets simplified in the case of either uniform damping or zero damping. For the uniform damping case, we have the equation angle as (3.24) for center of
Vo * Vo
n since Z D.O.
C0A
(3 27)
"
(3.21). i=l
1 1
Vo
C0A (3.25)
+ P
(3 28!
'
Vi * "Vi
C0A
i=1 2
'"
(3 29)
For the
zero damping case, we set D^ " 0 for all i. The state space model in 2n dimension space with the state variables as [w^,. . . ,u>n (J^ ,.. . , 6nJ for the uniform damping case is obtained from (3.29) as M.uj. = D.w. + P. i i 1 2  P . i   P_ ei M_ COA (3.30)
Equation (3.30) is decoupled trom (3.27) or (3.28) and for both uniform and zero damping cases the correct dimension of the state space is (2n2).
1
Analogous to the machine angle reference case, we can take the n*'* machine as reference machine and write the state space equations from (3.30) as 5  & n  0 P. P P . P  A<fi.t5 ) + {  ~&) <i M  ) M i n M. Mn i n (3.31)
0.
i i
= . n
 ui i
80
= 0.
It is not surprising that these equations arc identical to (3.18) since w.  w = n?.  tii and f).  b = 6.  6 . I n I n i n i the state space representation angle. equations 3.6 Tor
Thus,
the same whether we take the machine angle or center of angle as reference In the case of nonuniform damping, however, (3.25) rio not lend themselves to a neat representation.
State Space Models with PostFault SEP Transferred to Origin (Machine Angle Reference)
The preceding discussion has focussed on different formulations of the state space equations used by many research workers in the literature who have constructed Lyapunov functions based on energy considerations. Lyapunov stability work, we now cast the equations origin is the equilibrium point. discussed in Chapter II. As discussed in Section 3.3, we seek to construct a region of stability around the postfault stable equilibrium point. Hence, the differential equations that we must consider arc those corresponding to the postfault state given by (3.15). The
v re<i
For
matrix in (3.11) must be calculated for this case to If there is no line switching following a Let us the postfault SEP is the same as the prefault operating point and
the ?rej matrix is the same as for the prefault configuration. line switching. The equilibrium solution of the postfault system the equations (3.15)
consider the yeneral case when the two configurations are different due to
is obtained from
= 0 i=l,2,...,n
(3.32) (3.33)
are n equations in
differences.
Hence, we may solve only the first (n1) equations for the In general, we can write the constraint by summing (pP^ 1 ei
angle differences while the remaining equation can be interpreted as a constraining equation. up all the n equations in sine (3.33), i.e. I i=l terms in this expression cancel out, = 0 where P i xs defined by COA parameters pertaining to the postfault configuration. terms of COA notation, this implies 6Qs 0, in In i.e., (3.26), (3.23) with the Y.. P^f) = 0. Since the
81
the system regains the original frequency. zero transfer conductances, the condition V n n ? I P. = 0 , i.e., t (P . E. G..> mi 1 11 iml 1 iJ existence of a solution of Let the solution for Eq. (3.33) (3.33) be 6^.
In the case of
COA
=0.
if very often quoted in the literature. In solving (3.33), one of the We are seeking a stable
angles 5. generally the rotor angle corresponding to slack bus, is taken as reference and equal to its prefault value. operating solution and, hence 6? will be in the neighborhood of the
fi^ = fi?
(3.12), lEil2Gii
=
mi "
,
lEjlE^lY^lcostO.. *i
(S^)>
[C..sin(6f6!)
+ D..cos(uf6^))
(3.34)
' Pei<^ Hence, as d25. d6,  ~ + D. k = P . P 1 dt2 at2 = ^i  cos[0i^ For ease of notation, from E., E. and Y...x 1 3 3 * *S1
I
T
the
M.
1
. (6)
(3.35)
With these variables (3.35) can be reduced to the form x* = A x*  B* f(C*x*) This model is the equivalent of point. (3.16) (3.36) with origin as the equilibrium
3.6.1
The question has been investigated in the literature from different angles viz.
in Sec.
3.4.
However, all of these ways of looking at the problem lead to the same ultimate result, namely i) for the nonuniform damping case, i.e.
the minimal dimension of the state space is 2nl. ii) case, i.e.
D AM, Mi1 n
S M 2 J
D JL M.
D,
0 for all
The discussion is facilitated by assuming a model with zero transfer conductances. D^ ^ 0 in (3.34) do. + i and With this assumption, (3.35). Equations
= <i/2,
i.e.
IF
do. n = ^ PiEjH.j[sin(q^)sin(d.5j)l
jii
(3.37)
D A = M I
n 1
n 0
H_1K
0 B 
n m
n n n
c = r o
m =
n(n
^
83
D
K
Diaq(Di)j
M = Diag (M.)
n1 1 0 0 1
n2 1 1 . . . 1 0
n3 I 1 0 . . . 0,.. . I 0 0 1 1 1 1 0 1 1 1
1 l 0 0 1
1.. . 1 1 1 1 1 1 1 1 0
1
1
1
0 0
' I
i i
f
with The sysLem
SinS ]
3.4.
The
(3.39)
i ^ j) ii)
uniform damping A^
The zero damping is a special case of W (s) with X = 0 and the 1 T 1 ~u transfer function is WQ(s) = K M K. 3.6.2 Minimal Realization
Physically speaking, we would like to ensure that the number of state variables in (3.38) is a minimal set giving rise to the transfer function
matrix WN(s). C)
(A, B,
the minimal order of the state space model. state space model function '($') , of the form W(s) *
n
o*s*
as L ie caSG ma
'
has
[12]
that if W(s)
i1
I _1
1 Z.( S +A.)
1
If W(s)
 a(s)C with a(s) a rational scalar and C a constant matrix, then the degree of W(s) is equal to degree of a{s) times the rank of C where degree of a(s) is the highest power of s in the numerator or denominator of a(s).
85
Klx 3.5. Orlpnted ftVftpt) for the Internal node description of angle variables. Nonuniform Damping W^fs) can be rewritten as
where E\ is  .
is an nn matrix having all zero elements except the ii clement which The degree of WN(s) = Rank of
>:
(KAD1K) +
Rank of ( K L F
K)
7AK is
unity
K is n1.
therefore the minimal order of the state space for the system with nonuniform
constitute the minimal realization for the n machine power The following state variables spanning the
= w
dt
i=l,2,...,n
.... = n+i
(6.6(6?6 ) S) a n
i n
i=l/2,...,n
The state model is given by x = A x  B f (a) a 3.40) = C x n n3 DM"1 A = n1 n 0 0 (mn+1) It may be easily verified thai, the transfer function WN(s) computed with the above state space description is identical to (3.39). n1 I
T
0 B =
n1
The preceding discussion can be readily extended to the uniform and zero dampinq cases. We will only derive the minimal order of the state space. L7J. The minimal realizations are discussed in Ref. Uniform Damping
w fi u( )
ts
 V  K M
""K 
,  A V" * M
sA
( S +A)A
T 1 Since the rank of K M K is (n1) , the degree of W (B) is 2n2 and, hence, it is the minimal order of the state space. Zero Dampinq The transfer function matrix Wo(s) " K M s' 3.12) ats) KTM"XK T 1 Since the degree of a(s) of WQ(s) 3.0.3 = 2 (n1) = 2 and the rank of K M K  n1, the dcqree  2n2 which is the minimal order of of the state space. K is 1 ,.T..1
We will now illustrate the state space equations including transfer conductances for a 4machine case and then generalize for an n machine case with nonuniform damping. case. Four Machine System The state variables are x where The reason for choosing a 4machine system is that the structure of various matrices becomes clearer than in a 3machine
t V V V W4i614~614'624~624'634~*34jT
 4 (i=l,2,3)
64 = 6\
The formulation is facilitated by the introduction of auxiliary variables a^s as follows. 'l
y = 6
14 " 614'
5
2 ~ %4 " 624' 5
= 5
3 0
% "
fi
34
= 5
12 "
12'
13 "
13'
= 6
Note the manner in which the a.'s are ordered. the state variables x; rotor angles. (1=5,6,7)
The first
q=2,3,...
{q > p)
87
bus is connected to every other bus and bus 4 is the reference bus. Expanding derived as:
k
(3.35)
= Ax  BAf (o)
 2g(o) (3.43
o = Cx ) " Let A i = Di/Mi. nonlinearities f below: <X1*2X3X4!X5X6X7)T' The matrices A, 2 <X V ' j f ' V s ' V
?
h 0
0 0 ] 0 0 ^3*4
0 0 0 0 1 0
0 0
h 0
0 0 1 3*3 1 0 1 0 0 1 M 3 1 M 4
0 0 0 ~4 1 1 1
A
1

i
>
2 4*3
!
'<
2 3* 3
j
\ I 1 1 l o 0 1 0 J M 4
0 1 1 I M I I M 2 0 0
3*4 1 0 0 X M 4 i
M
0 1 M2 1 M 3 0
0 1 H 3 0
J
1 I 0 0 1 4
0 1 M2 0 1 4
0 0 1 3 1 M 4
M
1 i 1 M 2 0
M
1 l 0
0 1 M2 1 M3 0
M3 0
^l^
cos&l4]
E E B
3 4 34tin^3+634)
Sin^4l,
g3<c3)
i 2 ] 2^cos *4+612*
"
COSl5
i2^ 
= E2E3B23Lsin(c6+623) =
E E G [cos 2 3 23
, ^23]
gg(o6>
*a64'fi23J
89
nmachine System For the general n machine case, the state variables arc defined as i=l,2 ,. . ., n i=l,2,...,nl Define f^(c) = EEBr, [sin X (c^+<5>  sin6pq1 pq' p q pq s = EpEgGPgIcos(0i+6 For i=l,2,...,nl p=l,2,...,nl For i > n1
cos 1 pq
(3.44 )
a. (5pSg>  (S*^>
(3.45 )
D M
1
n B ^ n1
M"IK
n n1
m M^L n n1
n 0
n1 S i m n n1
0 J' 
in1 0 ... 1
1 0 52
=
0 1 o 1 1 1 1 1
L 1 ...
1 . 1 . 0 .
0 I
I
1
I
II
C =
1 0 1 1 K = 0 n
[ 5i ! 52 !
! J
5nl
o where I  Identity matrix, L has same structure as K with 1 replaced by +1. It may be verified that n1 mn+1 n i.e. T in 1 = I K ] n (3.46 )
5i
if the transfer poniluptaiiees are neglected, then g(a)  0 and the model reduces to (3.40) with B^ = B.
3"w4 j
,5
S 24~524'
fl
S 34~6341
We set = \j ~ ~ A4 "
91
= A X oh
 C x
 A
0
>
0 0 1 0 0 1
0 0
o 0 0 i
0 0 1 0
\
0 0 1
1
1
l
o
3 0
1
0 1 I
! 1
M4
1
M 1M
77
5]
1N
1
M
1
M2 M4 _l_
M
1 3 M 4
M ~
1
=r=
M,
M ,
1
3 M 4 1 M 4 1
M
J, M1
M
^2
2 " M4 1 J M
1 2 0
si _1
M,
1 M .
H.
 B 2 g(o) o =
(3. 4 8 )
ICAL MODELS
92
n1
T
1 0
1 
n1 n1
C =
5i *
*2 =
X=0
in
(3.48).
State Space Models with SEP Transferred to the Origin (COA Reference Frame)
For the multimachine system with the classical model, we finally consider the state space model in COA reference frame with SEP transferred to the origin. The procedure is analogous to that in Sec. 3.6. The state is space models developed in Sec. the equilibrium point. Honuni form Damping in the case of nonuniform damping, Eqs, considered. (3.24) and (3.25) are to be 3.5 are to be converted so that origin
damping terms and, hence, they cannot be cast in a state space form in any meaningful manner. Uniform Damping The equations to be considered are repeated below for convenience. D.
6
(3.27)
and
(3.29)
which are
T * ~ M^ !'o
(3.49) (3.50 )
P. l
1
93
The minimal dimension of the state space is 2n2 and we accordingly def ine the state variables as x =
V%'VSn'"'fr!nl^n!flln"9ln'e2n"82n'
'Lnl,n LnlfnJ
The state equations are obtained by first subtracting the n1"*1 equation in (3.50) as from the other equations and then expressing the resulting equations
u>
i"un
>'('"i^n)
<  M~
v
VT^]
P
. (e) M.
1
(9) _ (V"' _ M / n
^n  )
= fi. in is identical to
= ft. , it will turn out that Eq. Art For zero damping case X I 0.
Wc , therefore, conclude v.hat. for uniform and zero damping case, if we cast the equations in minimal state space, it is immaterial whether we take machine angle or center of angle as the reference. 3.8 State Space Models Including Flux Decay
One of the more realistic efforts at improving the synchronous machine model is not to assume the voltage behind transient reactance as constant. would then account partially for the flux decay effects [13]. effects, we shall assume that no longer held constant. expression for P^. Hence, Eq. now time varying. state be 
l ^
E S an
This
In the model which includes flux decay lE^J (i=l,2,...,n) E in (3.12) and (3.13) are
We will, however, assume that the transfer is held constant only in the applies except that E^[:,EJ. (3.37) becomes B^,E_J are in the postfault steady (dropping the (3.37) Then Eq.
<*
notation
dt
3"*
94
i=l,2,...,n
(3.52)
The change of the internal voltage of i*'*1 generator is described by the first order differential equation dE. n i  a. (E.E?)  3  2 (3.53) dt X X X X D. .K. (cosrl?.cos6. ,)
XJ J LJ J, J
for i=l,2, constants (sec Rcf. [13] for details). (3.52) and
. Fn where
a^ and 6^ are
l"EtEn<]T 
95
n n1 1 DM 0 0 A =
n
1 1
m .1 K
n
n n
$ 0
0 0 a
0 
"
n1 B = n
'~
0 o
i
11
0 n1 e
where
A is
(3nl) % <3nl)
matrix,
B is
(3nl)x (m+n)
and C is
(m+n) x (3nl) matrix, K and K are as defined in Eq. diagonal nxn matrices. n 0 C = i 0 0 n1 i n 0 n1
i
0 I (mn+1) n F(o) =
f
1 _ _ _ 1 0
f , (o) 2 n
li(^
" WVV^i^pq 5
" s^pq>
il^,...,m
i = l ,2
The above system describes a nonlinear system with multiple, memoryless, coupled nonlinearities. They do not satisfy the sector condition just as However, by putting them in form of the multivariable Popov is possible to arrive at
the
systems with transfer conductances do not. a Lure* type form and using a modified criterion [14] constructive procedure
Lyapunov function. Conclusion In this ehaoter, we have discussed several state space descriptions of an nmachine system with nonuniform, decay effects. REFERENCES; 1. Anderson, P. M. and Fouad, A. A., "Power System Control and Stability", (Hook), Iowa State University Press, Ames, Iowa, 1977. Staqg, G. W. and ElAbiad, A. H,, "Computer Methods in Power System Analysis", (Book), McGrawHill, New York, 1968. Pai, M. A., "Computer Techniques in Power System Analysis", (Book), Tata McCraw Hill (India), New Delhi, 1070. Willems, J. L., "A Partial Stability Approach to the Problem of Transient Power System Stability", International Oournal of Control, Vol. 19, No. 1, 1974, pp. 114. as well as uniform damping with zero as well as nonzero transfer conductances and finally a model including flux
2. 3. 4.
96
5.
Sastry, V. R. and Murthy, P. G., "Derivation of Completely Controllable and Completely Observable State Models for Multimachine Power System Stability Studies", International Journal of Control, Vol. 16, No. 4, October 1972, pp. 777788. Sastry, V, R. and Murthy, P. G., Discussion of Willems' paper, "Direct Methods of Transient Stability Studies in Power Systems Analysis", and reply by author IEEE Trans. Automatic Control, Vol. AC17, No. 4, August 1972, pp. 480581. Pai, M. A. and Murthy, P. G., "New Lyapunov Functions for Power Systems Based on Minimal Realizations", International Journal of Control, Vol. 19, No. 2, February 1974, pp. 401415. Tavora, G. J. and Smith, O.J.M., "Stability Analysis of Power Systems", IEEE Trans, on Power Apparatus and Systems, Vol. PAS 91, Ho. 3, May/June 1971, pp. 11381139. Stanton, K. N,, "Dynamic Energy Balance Studies for Simulation of Power Frequency Transients", Proc. PICA Conference, Boston, May 1971. Fouad, A. A. and Lugtu, R. L., "Transient Stability Analysis of Power Systems Using Lyapunov's Second Method", IEEE Winter Power Meeting, Paper No. C72, 1456, New York, Feb. 1972. Padiyar, K. R. and Doreswamy, C. V., "Analysis of Dynamic Equilibrium of Power Systems", Proc. 7th Annual South Eastern Symposium on System Theory, March 2021, 1975, pp. 123125. Gilbert, E. G., "Controllability and Observability in Multivariable Control Systems", SIAM Journal (Control), Vol. 1, 106263, pp. 128151.
6.
7.
8.
9.
10.
11.
12.
13.
Krfkimoto, N., Ohsawa, Y. and Hayashi, M., "Transient Stability Analysis of Multimachine Power Systems with Field Flux Decays via Lyapunov"s Direct Method", IEEE Trans. PAS, Vol. 99, No. 6, SopL/Oct. 1980, pp. 18191827. Desoer, C. A., Wu, M. Y., "Stability of a Nonlinear Time Invariant Feedback System Under Almost Constant Inputs", Automatica, Vol. 5, 1969, pp. 231233.
14.
CHAPTER IV
LYAPUNOV FUNCTIONS
4 . 1
J ntroduction
In the previous chapter, we developed the mathematical models of multimachine power systems used for transient, stability analysis. (sec. 1.2), In Chapter I it was shown that the critical clearing time ^CJ. can be computed This is accomplished by constructing a Lyapunov > 0 and v(x) < 0. If V(x)
from knowing the region of attraction around the postfault stable equilibrium point. function V(xJ inequality V(x) for the postfault system and defining a region Q by the < C so that inside S2,V(x) = 0 should not constitute a trajectory approximates and the value
 0 inside ft, then the points V(x) of the system excepting x = 0. of C which defines il. of stability.
theorems, the region defined by V(x) < C is contained in the exact region In this chapter, wc shall discuss the systematic procedures While for the simple to construct V(x). This is done for both single and multimachine systems using the different techniques explained in Chapter II. models, different methods may yield the same V(x), the methods will be useful when complex models arc considered. Techniques for computing the region of attraction are discussed in Chapter VI. 4.I Single Machine System
Consider a single machine infinite bus case whose swing equation is given by
M where P
at
+
A
D if  P  P
ot K E
sin 6
(4.1)
sin 6 = e
12
the internal voltage magnitude of the machine, E^ is the magnitude of the voltage of the infinite bus and G^ is the selfconductance at the internal bus after eliminating the generator bus where the load, constant admittance. bus. Equation if any, is converted into a is the reactance corresponding to will have the equilibrium points given by the
the postfault system between the generator internal node and the infinite
(4.1)
97
98
2
(4.2)
o = o
.  I P = ,u n  s in
Ps p = II ft
It is easily verified by linearization of (4.1) that 6s is the stable and 6U is the unstable equilibrium point. In tact, the unstable equilibrium point can be shown to be a saddle point. Eor Lyapunov stability analysis, stable equilibrium point 6  6.
s
(SEP)
Thus, M ^5 2 + D dt
j
e
(4.1) becomes
dt = P^sin 6s  p e  P sin (x+6s) c e sin fiS)
= (P Equation
s
sin (x+5S)
(4.3)
(4.3)
p siri e
is case in the state variable form by defining x^ = x = Thus Equation (4.4) has the origin x. The nonlinearity f(x^) has an (Fig. (Fig. 4.1). 4.2). The It is
6  6 , x^ = x = 6 = w. where f (x^ =
interesting property, namely that it passes through the origin and is in the first and third quadrant in a region around the origin nonlinearity f(x^) P = Pya i n 6 in Fig. points. 4.3 byapunov's Method and Equal Area Critcrion 7Metfiod~bf First Integrals!
s
easily verified that the intersection Of the horizontal line of constant P 4.2 occurs at 6s,
T
6s,
ii  i5s.
In this section, we provide a link up of Lyapunov1s method with the classical equal area criterion which is generally discussed for the zero damping case. We apply the method of first integrals discussed in Sec. 2.11. the swing equation (4.1) with D = M
0
Consider
4  P  P sin
(4.5)
2 (4.6 )
2  " af 2
f(X
1>
or.
99
2
Since R= + = 0, Eq. (4.6) has a first intcqral V<x.,x9)=C. From we obtain
dx
(4.6)
l dx2
2 (i/MjfncjT
(4.7)
Pesin6
Fig. it. 2.
IV
100
V(X X
T 2) ^ T
=
/ 1 mm
2 I
(P x
X
M ^0
sin (u+6Sj
Multiplying the righthand side of the integral, we get V(xx,x2) + Pg cos =  &xf  P X]L
GCOS
(x1+i!is)
Pg sin 6s
(4.9)
Since V > 0 and To  6 and
s
V = 0, the origin is stable in the sense of Lyapunov, relate V to the physical variables, substitute x^ = 6 = 5 = a; in Eq. V(<5 ,w) =
Mco
 PE cos 6 + Pe cos 5s
(4.10)
Since P = P
Mco2  Pe(cos6cos5s)
 P(6Ss)
(4. 11)
The constant Vcontours in the state space are shown in Fig. <]. 3. Since V(6,w) is also a first integral of motion, the + constant Vcontours are also the phase plane trajectories. Physically ~ Mu>2 = VR Kinetic Energy and [P(66S) P^(oosrtcosfi )1 can be further interpreted as the sum of positional energy P(c"6S) reactance servative , V(o*,ai) k  V, (tu) p (6U,0) + V (5) = constant of the rotor and magnetic stored energy in the PQ(cos6cos6s). Since the system is cons
Potential energy
(4. 12)
is
The separatrix passing through the saddle point defined by the constant Vcurve V(5,u0 = V(5 ,0).
U
The closed
region inside of the separatrix defines a region of stability. From Fig, (4.4) and Eg. (4.11), V(6U,0) m P(6U<5S) Pe(oos6 cos6 ) Area 2 + Area 3. During the faulted period d^*i5 f f *^'l^'2 M rr2 = P  P sin 6 where P =  , x = reactance between e xf ' internal node and infinite bus during the faulted period.
at
Area 1 = 6 =
101
since u
= 0
Hence,
(4.13 )
This is precisely the equal area criterion in the literature. With damping, the equal area criterion gets modified as discussed in Rcf. 4.4 Variable Gradient Method (sec. 2.13 of Chapter II) to Eq. (4.4). Assume a [lj.
llXl
rr
12X2 (4
X + a
VV =
VV
21 1
22 2 3VV2 .
V is computed as
V = "
* x2(ft12~a22
f(x J
D/MJ
21xl+ft22x2>M
(4.1b)
12
1 >e cur
* equation is
TV
102
21,v
(X,
 = cu,
3x
 12
"21
3a 21,k (4.16 )
a.
"1
0 also.
Letting
tX 2  l* = 
9e^
" 2
V = (11/M) x l2
(4.17 )
and
VV 
(4.18 )
TV
103
Multiplying the righthand side of (4.18) by M gives the same Lyapunov function as for the conservative system. not appear in the Vfunction. semidefinite function of x^, by choosing vt^ = 2 and n ^ 2 = ct^j = 2 ^.
2
D; (4.19 )
f (x x1+x2) D2
f x
V  xlX2(an  2
^)
 2
< l> Suppose we further set = 2 (7 x, + ) /x, then, Ctx^ M D V = 2 ^ x1  which is negative semidefinite in a region around the origin. D2 ^ l
X f(xl> +
2D
x
2 (4.20 )
W =
2D 0 IT Integration of VV gives
Xl + 2X
. 2
I'"
(4.21)
Equal
area
(4.22) If D 0, the above Lyapunov function reduces to the energy function derived The region of asymptotic stability is obtained us
previously.
105
V(x1,x2) < V(fi ,0) whe re rtu is the saddle point given by 6,u
=
T
 fs.
For single machine systems, there have been other approaches to the construction of Lyapunov functions such as Cartwright and Aizerman's Method [2], format method [2], energy metric algorithm method Mansour 14] L4], etc. All of these methods yield the same Lyapunov function as in the variable gradient method for the classical model. has extended the energy metric algorithm method to the single machine infinite bus case when the machine is represented by a third order model including flux decay effects. 4b Zubov's Method
[5]
(i) theoretically it gives the (ii) it is suited for computer
This method is attractive for two reasons; exact region of stability and
implementation if the nonlinearity is expressed as a power series. Although a decade ago, the computer memory requirements using this method was considered as a deterrent, it is no longer true now. Zubov's method has been discussed in Chapter i1 (Sec. The theory of 2.14).
We consider a slightly more complex model of the synchronous machine than in the previous section. Effect of transient
salicncy and nonlinear load damping is considered. The derivation of the model is contained in Ref. The swing equation is qiven by M [5].
^4 +
at*
D(6) ~ = P  P at 2
6
sin e
&  P0
sin 23
s (4.23 ) s x
+ a^ cos
2 6
, P , a,, a arc all functions of system and e machine parameters. The equation is normalized by defining
T
^4 + 2
dl
D*<6)
M = dt
(4.24)
where P e e e
Assuming that the equation pertains to the postfault state, the stable equilibrium point sin (SEP) <SS is obtained by solving
Defining the transformation x  6'  6S which transfers the SEP to the origin, Eq. (4.24) becomes ^ + D"<x)
+ R(x)
 0
(4.25)
106
dx D"(x) and R(x) composed of sine and cosine terms can be expanded in a power
...
(4.26)
R(x)
(4.27)
k?
= D''(x1)x2  R(x^)
Followinq the method discussed in Sec. 2.14 of Chapter II, the Lyapunov function is chosen as v(n,(x1,x2)  v2<xrx2) * y3 u 1,x2) ..+Vn(x^xjj where i.e.
v x m( i,x2)
I j=2 k0
E
JK 1
d..x? Kx*
2
(4.28)
v2(Xl,x2) = d^ + duxlX2 +
VW
d
dQ2x22
X + d
" 30 l
21 l 2
X X
+ d
12 l 2
X X
+ d
03 3
(4.29 )
coefficients arc determined recursively as discussed in Sec. 4>(x^,x2) in the p.d.e. Q. f5l in
0, 6 > 0, aft
(2.24)
Numerical Example
A synchronous machine is connected to the transmission system as shown Fiq. 4.5. p.u. and the initial terminal voltage of the machine is 1.05 p.u. A. Line AB is cleared at 0.0833 sec. stable.
3phase to ground fault is placed at the transmission line AB near the bus and after the fault is cleared, the line is reconnected to the system at 0.4 sec. Investigate if the system is The followinq data and system differential equations pertaining Voltage behind transient reactance = the dimcnsionless time
T
The swing equation during the faulted state until t  0.0833 sec. ) The swing equation for .0833 < t < .4 is 5 
is 0.6489 (4.30
^Z +
(0.03072 cos2x + 0.01332 sin2x) ^ + 0.7258 sin x d i  0.07227 sin 2x  .6489 (4.31)
IV LYAPUNOV FUNCTIONS
107
.4 sec.
is
(0.04972 cos2x + 0.02468 sin2x) ^ + sin x  0.1291 sin 2x  0.6489 (4.32) into the state space
(4.32)
ST
dx2 dt
= *2
= D"(x1)x2  R(x) The power series expansion of D'"(x1) and R(x^) is obtained as
DQ
+ D1x1
= 150 +
ji
1<U34)
R(x,) =
(cos(*s + e=i *
j.
* (4.35) and
30.
The approximate stability regions computed from the tor 4.6. The function <t>(x^,x.j)
a = 6  D Q  (1.0372 + 0.0125 cos 26s The region of stability increases with n initially for n  5,8,10 and 16 but decreases for n  26 indicating that the approximate region of stability does not approach the true boundary monotonically with increase of n. over that of n = 16. particular example. this method. in Sec. The stability region for n = 35 was computed but indicated no improvement The best value of n depends on the This constitutes one of the drawbacks of is determined. the fl.. (81 As explained first until
Figure 4.7 shows an example of how the boundary coefficients are are searched
determined.
'(x) = 0
108
is found.
denoted 2
different choices of 0(x^,X2), the one with 4  DQ {x^tx^) results with C  0.5057. To ascertain the stability of the system, (4.30) and final switching.
(4.31) are integrated from the instant of fault until that of For a fault clearing at t = .0833 sec. and final switching although outside of V ^^
at t = 0.4 sec, it is found that at t = 0.4 sec, the state of the system falls inside the stability region V ( 1 0 J , V * 1 6 1 , and v ( 2 6 j and V^^. gives only sufficient conditions . The extension of the above method for the case when flux decay effect is included is also possible.
M
&y 2 +
dt
D( 6 )
~  P  P dt
H
(4.36)
dE' dt ex q
5
= K E___  L
El +
M CO:
109
0 X 0.013 P. U .
\ i X
 _ c 
X (Y/2) /2)
(Y
u
4 x
(Y/Z) machine
Single
INFINITE P BUS Q
Y/2 n
OF THE SYSTEM
^3
===^
N = 5
^4,//
//
to
III ( (
\\\\ \
/
/ /

) ) /)ft / s'Ss'X ^
////
1
1. 4.6. by 0 of n
x: FINAL SWITCHING I l 1.0 0.5 0.0 0.5 1.0 SLability boundaries Zubov's method for different val

V{8) = O's*
1.0
V(8) > 0
0.5
0.0
0.
y v
o ^/Z^^W = o
(s) >
110
\ J)
1.0
1.5 Fig.
where K, L and M are constants depending on machine and system parameters. Although the procedure is same as in the previous For details, refer to Ref. [6]. example, programming requirements for computer implementation are considerable 4.6
As discussed in Sec. 4.2, the nonlinearity for the sinqle machine infinite bus case lies in the first and third quadrants in a region around the origin things: stability, (i) (ii) (Fig. 4.1). This implies two we can apply Popov's theorem for absolute the absolute stability property will be
modified so as to have a region of attraction since the nonlinearity violates the sector condition for outside the interval of (i) %2 < o < fej Uj > 0 and
< 0) .
in view
we can construct systematically the Lyapunov function 2.15.4. It is also possible to get an
via Popov's theorem and Kalman's construction procedure as discussed in Sec. analytical expression for an estimate of the region of attraction from a knowledge Jl^ and
t^,*
The state space equations (4.4) are rewritten in the form x == Ax + b f (o) T a = c x
i
(4.37 ) 0 1
where A =
1 b = 6)
S
c = 0
1/M = P sin(a+S e
" P
(4.38) =
The transfer function of the linear part is given by (2.44) as G(s) cr(sIA)_1b which yields G(s) = s(S+D/M)
The nonlinearity (4.38) is shown in Fig. 4.1 and is seen to satisfy the sector condition 0 < f< * in the region ( it26s) <
<
(ir26s) .
Since
111
Eq.
(4.37)
(4.39 )
S = 4>(a)
f {$) .
^*
0 0
+
0
1 (4.40)
u = < (o)
Cf
Re
t
jiisq
(4.41 )
i.e.
>
i)
n > 1.
4.6.1
Kalman's construction procedure explained in Sec. 2.15.4 i: now applied. For this system iji(s) = s + Step 1: Step 2: From Popov's criterion q * ~ where n  1. C<3 " 1' W (hi) ^  ^ M n * n M  *#) = "  1.
w
Step 3:
112
Factorization of WU) = 6(jw)6(jw) yields trivially e(ju>) Step Step Steo Step 7 4 5 & /n1. r = q(d+cTb, v(z) = 0. * /n1.
= 0 (z)
u  /n1 . The Lyapunov matrix equation corresponding to ATP + PA = uuT becomes  jjj p^j + p,^ ( ^) where P = p.y>
5From (2.55)
=  (n1)
we get,
v(x,f)  XTPX +
DC
+ q /
<MO)DO
2gr x2 +
K2
+ 5}
/Q
F(0>D0
g
x2
5 ^
/
ttodo
(4.43)
This is a fairly general expression for the Lyapunov function. Different choices of n yield different Lyapunov functions. D n = g with 0 < & < D gives Mansour's Lyapunov function [4], The choice of n = 1 leads to V(x1,x2J =  xx + x?)2 i f r (o)do (4.45)
which is equivalent to (4.22) obtained by the variable gradient me Lhod. To compute the region of attraction, we use the method due to Walker and Mcl.arnroch 18],
Consider the system (2.43) x = Ax  b$ (o) T a. = c x whose V(x) is given by = X/PX + q (2.54) as
V(x)
/Jj (t(o)do
I]
1 ^2'
Since the nonlinearity $(u) violates the sector condition at M^ = Min V(x) (1=1
r
define (4.45)
2)
113
T subject to
e x
Note that this minimum takes place at a point xn at which T ~ T c == j and VV(x^) is orthogonal to the hyperplane c x ^ (i=l,2).
1 T 1
s In the power system problem, ^ = T  26 , ?,2 = n  2$ hence  ^ ! < I Therefore, by inspection, the region of attraction is V(x2,x2) where M. = c P c * q (4.46) 'Q"2^ < Hx
*(a)da
We now apply this result to (4.37) whose Lyapunov function V(x) is given by (4.43). P is obtained from (4.43) as
D 2 Mn
p =
H 2 2n 1 4 n' (n1) D2
D 2n"
H 2
D_ 2n D 2Mn
T 1 P
(Tt2 n 2M
(4.46)
(4.43)
and (4.38)
as
'u"25
e(
ain (,+AS
)*sinflS)da "
attraction is given by 1 D 2 MK Xl 2
1
. M 2 7 *2 + n X1X2
, D 'o
/0.,,
f (o)d(J
<"268)2 n2 f tn1).
? '
[2P
cos 5s  P(n268))
(4.47 )
114
M x
+ r
f(o d0 <
>
<2PGcos6S  P( J 23S))
(4.48 )
It may be verified from (4.13) that since 6U %  fis, V(6U,0) = P('26S) + 2P cos 6s (4.49 )
Hence, inequality (4.48) is equivalent to the equal area criterion and the region of stability obtained by the method of Walker and McLamroch [8] is equivalent to obtaining C by evaluating V(x) at the nearest unstable equilibrium point. This aspect was utilized in much of the work since 1965 in obtaining region of attraction for multimachine power systems. The methodology using Kalman's procedure as well as constructing the region of attraction has been extended to higher order systems such as simple governor representation as in Ref. 4.7 [71.
The various types of models for multimachine power systems have been discussed in Chapter III. for these models. of application, Now we shall discuss how to derive the Lyapunov functions Most of the methods discussed for the single machine
cases have been extended to multimachine cases but from the point of view two tyj>es of Lyapunov functions which are general for the These are (i) energy type 4.3 but (ii) Lyapunov multimachine case seem to be of interest.
Lyapunov functions which arc analogous to those discussed in Sec. including transfer conductances but no mechanical damping, functions based on the multivariable Popov criterion. 4.7.1 Energy Type Lyapunov Function 110]
Although this function has been derived in many different ways since originally proposed by Aylett 19], we will adopt the COA formulation discussed in Sec. 3.5 of Chapter III. The COA formulation as opposed to the machine angle formulation has the advantage
that the
function can be given physical meaning analogous to the single machine case. Neglecting damping, the equations are given by = 0 (i=l ,2,...,n)
Vo
'
M
COA
<450
i''i  ei 
*i '
COA
I4
51)
*****
 j4 p i  2 j x .J +J
hf".q
(4.51) by 6^ and form the
115
.y A n n
M.
p + p
ei+ ^
coA)ei
w *i
p^.
* n
1
we obtain
T.
n1 6, 
Z
i
C J
sine
i
9 ^
i=l j=i+l
i i=l j=i
D
l
cose
x
n
1
i
i==i
z
j=i+i
coso
+O J ij i
TO
c44)
3
Integrate (4.52) with respect to time using as a lower limit t = t. where iii(t ) S 8 point. This yields
0 and
0(t.)
n1 I 1=1 j = m
OI+O.
[C,.(cose..cosOs,)
13
deriving a first integral of motion for the swing equation discussed in Sec. the contribution of the last term is zero since n M.u.
(4.55),
Ln.t6r.nal buses i and j, it is the generally held view that Hence (4.55) as it stands is of limited value. However,
transfer conductances prevent the construction of an analytical Lyapunov during the faulted state since we are integrating the faulted equations, we know 6 and hence the integral term in (4.55) can be computed using Lhe trapezoidal rule. The Tn limitation of the Vfunction given by (4.55) subsequent work, we shall refer to C as V terminology. instability, in Eq. arises from the need to compute to conform to the prevailing
C which defines the region ot attraction for the postfault system. One of the methods of obtaining V"cr is to compute the i.e. V
Vfunction at Lhe unstable equilibrium point corresponding to the mode of = V( U U,0). Hence, the upper limit of the integral to 6
s
(4.55)
is not known,
116
(This aspect will be discussed fully in Chapter V.) approximation for V^. can be obtained. definiteness of V and V.
Hence, an approximate
of one of the terms in V, it is not possible to judge analytically the sign From physical considerations such as the However, as will existence of the postfault SEP, one can argue that V > 0, but the same physical arguments cannot be extended to say anything about V. the energy integral or simply the energy used in actual large scale power i.e. V = i; t i e r gy Function Neglecting Transfer Conductances be pointed out in Chaj>ter v, this particular form of Vfunction, also called funcLion, has been successfully 110] and in the system examples
1
1
2
1
It can be shown that this Vfunction is equivalent to n1 n  V = E [~ M.M. (OJ . O ) . ) i*l }=+l 2MT 3 1
I
1
* (P.M.P.M.H^.o*.)
 Ci.cos(6ij6.)]
(4.57)
This Vfunction has been used by a large number of research workers. Aylett called it the energy integral by DiCaprio and Saccomano [30]
s
and RibbensPavella
also shown that V is indeed a Lyapunov function satisfying V > 0 and V = Hence, the postfault SEP 9 is Lyapunov stable. mechanical damping, we can indeed show that V < in a region around 8s is ensured since V / 0 on a trajectory inside that region. V > 0 in a number of ways. Mathematically speaking, V represents the energy integral, One can verify From physical considerations since it has to be greater than zero. In the presence of
quadratic term and the sum of integrals of nonlinearities, each of which lies in the first and third quadrants in a region around the origin. V > 0.
117
POWER SYSTEM
STABILITY
4.7.2
A more systematic way of developing Lyapunov functions by the multivariable Popov criterion is presented in this section. As a special case, the energy type Lyapunov function of the previous section. of the mathematical model discussed in Sec. convenient to use. it includes The Lure form
developed analytically only for systems with zero transfer conductance. With nonnegligible transfer conductances, there are results which take them into account only partially. We therefore consider first the system wiLh zero transfer conductance with nonuniform damping. System with X.ero Transfer Conductances model is given by Kq. (3.40) as (nonuniform damping) The state
x = Ax  Bt (g) a = Cx n n1
KT
(4.58 ) m
0 0 B =
where
A 
i
(3.44) as =
= fk(,k)
V :qBpqiSin(^pq>
'
sin
M)
{4 591
pq
pq
ion
^k'ck'k which we
>
'
<
kl'
kl
>
0,
k2
< 0 B (Fig.
The transfer function matrix W(s) is given by W(s) = C(slA) 3.4). Z(s) To apply the multivariable Popov criterion of Sec. = (N+Qs)W(s) is positive real.
2.16, more
popularly known as the MooreAnderson Theorem, we must find N and Q such that We shall take both N and
to be
The conditions for Z(s) to be positive real arc [15): (i) Z(s) has elements which are analytic for Re(s) > 0. (ii) Z* (s)  Z(s*). (4,61) T * (iii) Z (s ) + Z(s) is positive semidefinite for Re(s) > 0. The first two conditions clearly hold. (iii) holds. W(s) From (3.39)  i [KT(sI+A)~iM~1K] (4.62) Wc need to find n and q such that
119
POWER SYSTEM
STABILITY
where A = M *D.
W(s)
W(s) = i KT(sM+D)*"1K where (sMio) becomes ZT(s*) + ?.(s)  Z(s*) + Z(s) = 2Re[Z(s)J = 2K'''lRe(sns^ ' is a diagonal matrix.
(4.64)
1*
> 0
The at>ove condition will be satisfied if n and q are chosen such that the diagonal matrix "^sTsM.+D.)'1 
for Re(s) >
(4.65)
This clearly requires each term to be nonnegative and a sufficient condition on n and q is M. ta > n ^ Equation (4.66) also follows from the fact that X is 119). the is its
(4.66)
incidence matrix of an nnode passive network and W(s) admittance matrix satisfied in the MooreAnderson Theorem,
choice of N and Q does not cause polezero cancellation in M. W(s). This requires that ^ ? * or
= j.
However,
as we shall
123
see later this condition is relaxed in many cases. Hence, a M. valid range of n and q is q > n for all i and both q and n i are finite. Note the similarity of this condition to that derived in Sec. n with n > 1. [15] a 4.6 for the single machine case, namely q >
With N and Q chosen, there exists by the MooreAnderson Theorem Lyapunov function of the form Cx V(x)  xTPx + 2 fT(o)0 do *
(4.67)
A P
+ PA 
L L
PB = CTN + ATCTQ  L W
(4.69) (4.70)
where L and WQ are auxiliary matrices. model, it can be verified that CB = 0. Equations (4.68)  (4.70) to be zero. 9 = (li T A P + PA = L T/ PB = nC T +
qh C
(4.71 ) (4.72 )
T T
Let
N
n1
4
P ^
P
L n1
n1
2
T T P1 + P^Kj + KiP2
IT DM P2 ^ T I T
1 "?2
1
T T
3l
EiM~1k = <*K
(4.74)
121
(4.75)
(4.73).
21 * i2 "
"EiS" fi " SB
?2_1Eueations From (4.77)
+
*
2
~ii 
(4.76) (4.77)
31 to
(4.74) we get
(4.77)
P^, P^ as follows:
(478)
(M'^ H"3 
)K
= 0
(4.79) in each
K has the structure of an incidence matrix and has only +1 and 1 of its columns.
1
Therefore (4.80)
where U is a unit matrix with all entries equal to unity and r is a constant. This yields = 1 qM ) Now eliminate P2 from 131"1Hence
= n
(4.75)
and
(4.82)
*SH5
(4.83)
where s is a constant. It may be verified that Eq. (4.83) has the solution P _ = nD , + sD ,OD , 3 n1 n1n1 (4.84 )
is the matrix
with n'"*1 row and n1"*1 column deleted. and (4.78) respectively, the matrix
Substituting P^
2qD
since V ^ is symmetric. T Substituting the expression for K^P^K^ from (4.83) into (4.85) W H get 2(  q l ) l r i M ) (rs)
(MUD+IJUM) <
Divid ing by
122
(4.86 )
2(D F(y>
i
 M)
 ; ( M U D + D U M )
<
.e.
< 0
is negative semidefinite.
M.
f >
fD M
<i :i'
M
2n
(D
+ n
>M

i^l i=l
M
a  2
Z i1
(D,
i n
(4.89 )
satisfying det(F(u)>
 0.
= 0 can
M.
 a z
i
/ DiMi M. 1=1 \
D.
 n i
 1 0
(4. g o >
Different choices of n and q satisfying the positive real property and different choices of p lying between 0 and will result in various Lyapunov functions. (4.84) (i) The matrices P^ P_ 2 * P 3 are given by (4.81), (4.78) and respectively. Lyapunov Function of ElAbiad and Nagappan Choose n = 0, q = 1, 52 r = 0, s = 0. [111, Gless [161
 E3  0.
V(x) 11 x
(4.91 )
1 1
1
X
0

x + 2 /" f (a)do
.
Dividing the righthand side by 2 and reverting to the physical variables 1 V(x) = ^ f;(u,)du. (ii) +
7
m
fQ
i (4.92 ) [12] where y.j is the solution (negative) This results in the Lyapunov function
i=l
I 32
M + 1 V(x) X 
MUM
o
n j
/Q f
2 Tx + 2 (u)du
(4.93 )
4
2
(u,)du
(4.04) "i"i
/n1
f.
(i i i)
116J
n I
r =
0,
ZD . 1
3,
2i "
N
1=1 1 . where D n1 , is
From
 =L D
.bTJ n1n1
(i1,2,...,n1)
Ea^i?1^
2
+
V( J 2
 ill fil ^ ll
, n1 _i_ n .
y
r
I*
*S
2D
j=I+l n
m (iv)
I 32
4 m
i=1
^ g
i 1
j1 (4.96)
o. f. (u.)du.
l 1
(v)
[19J is obtained
V(x)
. = 
n E i=1
+ i
*

t
1
n E
n D.
M. D.Dj(6  6* * m + E i=l
M
5  fi.
^  ^ ^
. /A1 f <u.)du.
0 1 1 1
(4.97)
The state space model is obtained from Eq. (3.48) by neglecting the transfer eonducLances, i.e. BE
0.
IV LYAPUNOV FUNCTIONS
x = Ax  B f (g,1 o
I
Cx
I n r.
(4.98 )
S
,6
In In' where
' n1 , m
ii
n1,n
nl ,h
n1 A =
2s ! 9
1
1
L
n1 0 C =
<>
n1
n1 I
n1
PA = L L
ism
'.rp
(4.99)
PB  nC
+ q A n1
n1
n 1 T
n1
El
2
P = n1 L 
E
3
The equation
(4.99)
becomes
 m m
+ p 1 X  X ?1+ A + P 3 6i i ^2 ii
A Pi + P, 2  2 3 0
' !]
;^
(4.100) (4.101)
SiSl??"1? =
127
P^M^K = nST
(4.102)
"Bl
P3
"
* ?1
2 + 2 
(4.103) (4.104)
 P2
I
(4.101) by M ^"K^ and rearranging, we get
Premultip3ying
(l"%EiSfS"* "
Therefore ~ qIt can be shown that
qM1)K  0
(4.105)
+ r

(4.106)
El " <%f
+ r
14 1071
"
where M . is the M matrix with n***1 row and n*n column deleted. n1 We can also show that P  nM 2 n1 P 3  A P2 . + sM ,;UM n1n1 . (4.108) (4.109)
Particular choices of n, q, r and s yield different Lyapunov functions used in the literature. (i) Choosing n = 0, q  1, r =  j^S s = 0 yields the Lyapunov function appearing in the works of Aylett [9], RibbensPavella [13,18], Willems [12]. Then P 2 = P3  0 T Denoting the relaLive velocity subvector x^ = (to. ..(,) _, V(x) by 2
v (x ) =
t W m E
M M
T i=l
m
j=i+l o.
n ii
ii
(ii) Ref.
IV LYAPUNOV FUNCTIONS
m :$
o. J'^1 fi(ai)doi
(4.111)
This Lyapunov function may be considered as the nmachine analog of the Lyapunov function for the single m/c case (Eq. (iii) n = A , q = l , r  s
4. 1/MT.
4.44) .
I
) }2
^
1 1
j=l
j = itl
1
4 (iv)
111
i=1
o. / u
f.(u.)du. a i
(4.112)
n  X , q = 2, r =  rp, s  T T The resulting Lyapunov function will have the quadratic part which is the sum ot the quadratic parts in (i) and (iii) and m c. the integral term equal to I f 1 f.(u.)du.. This Lyapunov u 1 1 1 i=l function was proposed in Ref. [16). (v) The most general Lyapunov function for the uniform damping case will be [19]
V(
*>
n
1
b: E Y
(4.113)
for 0 ^ a A. This yields the result result (iv) (i) for a = 0, result (iii) for a = A , and
for a = ^.
Zero Damping The state model is the same as for uniform damping with A = 0 and hence the Lyapunov functions are readily obLained from the uniform damping case by letting X = 0 . 4.7.3 Lyapunov Functions with Transfer Conductances has been derived in Chapter III {Eq. 3.45) for
nonuniform damping case as x  Ax  B^ cr Cx The components of f (a), an interval i.e. f^ (a^) satisfy the sector conditions in f(o)  B2 g(g) (4.114)
around the origin whereas the Lerms g(a) due to the transfer
129
conductances *3o not satisfy such a property. multivariable Popov criLerion is not possible. to define the nonlinearities
the difficulty have been reported in the literature (20,21,221. One way
W  w w
Equations (4.114)
x DJ
Ax
ID.
h(o) V )
Cx (4.115
Equation
(4.115)
now has a Lure structure where both h^(o^) and Gudaru [23] tried to apply the (4.115)
(oV)
to be in error by proving that it is not possible to find an N and to satisfy the theorem. Hence, there have been a number of efforts at approximating Lhe effects of transfer conducLonces One of the methods which results in x = Ax  ftl h(o) & = Cx If the term (B^B^) c[ (<J ) MooreAnderson Theorem. V(x> h(2> = xTPx + = hitoiJ = E Y 1 [sin(o. + 5S +0 p q pg * i pq ) sin(6s +6 pq' pq )1 pq' which destroys the Lure structure can be is amenable to analysis via the The Lyapunov function will be (t^B^glc) (4.116 [25] [2527J. from (4.114)
J'q
h(u)Qdu
(4.117)
fi(oi) This nonlinearity satisfies the sector condition 0 <  < in the interval  2(6^+6.^) 2(^q+6pg).
The other method is to have an over description of the system in an n(nl) state space and then neglect some G^j's [26]. Still another method is to use an approximation to the energy Lype of Lyapunov function 127]. V(x)
in Lhe meLhod involving energy functions and COA formulation, [10] is retained in Lhe form of a path dependent integral (4.55). is computed depending on the faulted trajectory and using the trapezoidal rule. In computing Vtx1**) This meLhod a linear path angle is assumed between 6R and 9U. will be discussed in detail in Chapter V. The
integral
IV LYAPUNOV FUNCTIONS
FOR
POWER SYSTEMS
4.8
Lyapunov Functions for Muitimachine Systems Including Flux Decay Effects J28] in the classical model of multimachinc systems, the internal In this section,
So tar,
voltages have been assumed to be of constant magnitude. to neglecting the effect of flux decay in the machine. framework of the MooreAnderson Theorem. in Sec. 3.8 of Chapter III.
This is equivalent
we will relax this assumption and construct a Lyapunov function within the The mathematical model is given
The transfer function of the linear part of the system is given by W(s) = C(SIA)_1B W2 (s) W2(sj (4.118) T KJ(s(sI+A)] 0
 1  1
M JK
I(sl+a)"1^
We use a modified version of the MooreAnderson Theorem due to Desoer and Wu [29] for the stability criteria. in
The system to be considered is of the form given by Eq. (3.54) with W(s) Eq. (4.118) as the transfer function matrix.
and F(o)
(iii)
0 for
If we can find N and Q to satisfy MooreAnderson Theorem with the above restrictions, then there exists a Lyapunov function of the form V(x) = xTPx + 2 \fy id) [28J. The theorem was
The proof of this theorem is contained in Ref. used in Ref. [311 machine system with flux decay effects. We shall now proceed to construct Lyapunov problem in the multimachine case.
For the system to be stable, N and Q have to be chosen so that Z(s) = (N+Qs)W(s> (4.119)
is positive real. N is chosen as N  0 for Condition (ii) to be satisfied because T i * F(o) No can have neqative values around the origin for a
131
is chosen as
n1
= Z Z
itl
The partial derivatives of V^tc) are written H S follows: ?/ck = B. . [E.E.sin(o\+6s.)  E?sfsin^?j = f( (o) 13 x 3k 13 I T 13 for k=l (2,...,m
9V
lk (4.121.)
1 9ok
T.
j=i
(4.123)
IV
Hence
Q  I.
To solve tor the P matrix, we basically follow the same procedure as in Sec. 4.7.2. n In Lhis n instance, the P matrix is given by n 0 2nl
r
1
L. ~
n1
n 2nl
i
0
l
P = n1 ^2 n 0
0 * _
=

1
4 n
?c
(4.124)
With the choice of N  G and Q = I, the matrix z(s) is expressed as KT(sI+A)"1M~1K Z (s) 0 s(sl+a) 1 &
Zjts) 0
0 (4.125) X2(s)
Since Ms)
and Z^fs),
to be positive real can bo tested independently for Z^(s) Clearly, Condition (4.61) satisfied, are satisfied for Z^(s) and 2j2(s). For Property
Zi(jw)
Z2 (jw) 
+ 22(ju:)
Diag.
It is verified that both (4.127) and (4.128) are positive semidefinite. Therefore Z^(s) and Z
2
Z(s) is also positive real. The solution for elements of P matrix in (4.70) (4.124) using Eqs. (4.68) 4.7.2. Because of the 
structure of P, we can solve for P_c as for Lhe classical model wiLh no transfer conductances. (4.91) Any of the quadratic forms in the Lyapunov functions can be chosen. P_4 corresponding to W2(s) is obtained as (4.129) (4.97)
133
<x)
jIxj
m
Pc x 1 o,
4
T AE
a 8
aE
ik<V "k
. . . ,EnE;J]T
(1
where
AE =
xx = A special
 ^ni
it affects the quadratic part only. Conclusion In this chapter, an attempt has been made to derive Lyapunov functions in a systematic manner both for single and multimachine systems using many of the techniques discussed in Chapter II. For multimachine systems, we focussed on the energy type Lyapunov function which was derived using the COA reference frame and the general type Lyapunov function using the multivariable Popov Criterion. The laLLer is seen to be more general and flux decay effects as well. can be extended to systems containing References 1.
Willems, J. L., "Direct Methods for Transient Stability Studies in Power System Analysis", IEEE Trans. Automatic Control, Vol. AC1G, No. 4, August 1971, pp. 332341. Dharma Rao, N., "Generation of Lyapunov Functions for the Transient Stability Problem", Trans. Eng. Inst., Canada, Vol. 11, Rep. C3, October 1968. PeczkowsXi, J. L. and Liu, R. W., "A Format Method for Generating Lyapunov Functions", Journal of Basic Engineering, ASME, June 1967, pp. 433439. Mansour, M., "Stability Analysis and Control of Power Systems", in: Handschin, E. (Editor) "Real Time Control of Electric Power Systems" (Book), Elsevier Publishing Co., Amsterdam, 1972. Yu, Y. N. and Vongsuriya, K., "Nonlinear Power System Stability Study by Liapunov Function and Zubov's Method", IEEE Trans., Vol. PASlG, No. 12, Dec. 1967, pp. 148085. He Sarkar, A, K. and Dharma Rao, N.r "Zubov's Method and Transient Stability Problems of Power Systems", Proc. IEE, Vol. 118, No. 8, August 1971, pp. 10351040. Pai, M. A., Mohan, M. A. and Rao, J. C, "Power System Transient Stability Regions Using Popov's Method", IEEE Trans., Vol. PAS89, 1970, pp. 788794. Walker, J. A. and McLamroch, N. H., "Finite Regions of Attraction for the Problem of Lure", Int. J. Control, Vol. 6, No. 4, 1967, pp. 331336. Aylett, P. D., "The EnergyIntegral Criterion of Transient Stability Limits of Power Systems", Proc. TEE, Vol. 105C, 1958, pp. 527536. Athay, T., Podmore, R. and Virmani, S., "A Practical Method for Direct Analysis of Transient Stability", IEEE Trans., Vol. PAS98, No. 2, March/April 1979, pp. 573584.
2.
3.
4.
5.
6.
7.
8.
9
10.
134
11.
ElAbiad, A. H. and Naqappan, K., "Transient Stability Region of Multimachine Power Systems", IEEE Trans., Vol. PAS85, 1966, pp. 169179. Willems, J. L., "Optimum Lyapunov Functions and Stability Regions of Multimachine Power Systems", Proc. IEE, Vol. 117, No. 3, March 1970, pp. 573578. RibbensPavella, M., "Critical Survey of Transient Stability Studies of Multimachine Power Systems by Lyapunov's Direct Method", Proc. Ninth Allerton Conference on Circuit and System Theory, October 68, 1971. Pai, M. A., "Transient Stability Studies in Power Systems Using Lyapunov Popov Approach", Proc. 5th World IFAC Congress, Paris, 1972, Paper 31.5. Anderson, B.D.O., "A System Theory Criterion for Positive Real Matrices", SIAM Journal (Control), 1967, Vol. 5, pp. 171182. Pai, M. A. and Murthy, P. G., "New Lyapunov Functions for Power Systems Based on Minimal Realizations", Int. J. Control, 1974, Vol. 19, No. 2, pp. 401405. Willems, J. L. and Willems, J. C., "The Application of Lyapunov Methods to the Computation of Transient Stability Regions for Multimachine Power Systems", IEEE Trans., Vol. PAS89, 1970, pp. 795801, RibbensPavella, M. and Howard, J. L., "Practical Considerations on the Transient Stability Power Systems Studies by Lyapunov's Direct Method", Revue E, Vol. VII, NO. 5, 1973, pp. 167169. Willems, J. L., "A Partial Stability Approach to the Problem of Transient Power System Stability", Int. J. Control, Vol. 19, No. 1, 1974, pp. 114. Henner, V. E., "A Multimachine Power System Lyapunov Function Using the Generalized Popov Criterion", Int. J. Control, Vol. 19, No. 5, 1974, pp. 969976. Pai, M. A. and Murthy, P. G., "On Lyapunov Functions for Power Systems with Transfer Conductances", IEEE Trans. Automatic Control, Vol. AC18, No. 2, April 1973, pp. 181183. Uemura, K., Matuski, J., Yamada, J. and Tsuji, T., "Approximation of an Energy Function in Transient Stability Analysis of Power Systems", Electrical Engineering in Japan, Vol. 92, No. 6, 1972. Gudaru/ U., "A General Lyapunov Function for Multimachine Power Systems with Transfer Conductances", Int. J. Control, Vol. 21, No. 2, 1975, pp. 333343. Willems, J. I.., Comments on Rel*. No. 1, 1976, pp. 147148. [23] , Int. J. Control, Vol. 23,
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24. 25.
Pai, M. A. and Varvandkar, S. D., "On the Inclusion of Transfer Conductances in Lyapunov Functions for Multimachine Power Systems", IEEE Trans. Automatic Control, Vol. AC22, No. 6, 1977, pp. 983985. RibbensPavella, M., "Transient Stability of Multimachine Power Systems by Lyapunov's Direct Method", Paper 71CP3 7PWR, IEEE Winter Power Meeting, New York, Jan. 1971. Kakimoto, N. , Ohsawa, Y. and llayashi, M., "Transient Stability Analysis of Electric Power System via Lure Type Lyapunov Function", Parts I and II, Trans. IEE of Japan, Vol. 98, No. 5/6, May/June 1978. Kakimoto, N., Ohsawa, Y. and Hayashi, M., "Transient Stability Analysis of Multimachine Power Systems with Field Flux Decays via Lyapunov's Direct Method", IEEE Trans., Vol. PAS99, No. 5, Sept./Oct. 1980, pp. 18191 827. Dcsocr, C. A. and Wu, M. Y., "Stability of a Nonlinear TimeInvariant Feedback System Under Almost Constant Inputs", Automatica, Vol. 5, 1969, pp. 231233.
26.
27.
28.
29.
IV LYAPUNOV FUNCTIONS
135
30.
Di Caprio, V. and Saccomano, F. , "Nonlinear Stability Analysis of Multimachine Electric Power Systems", Kieherche Di Automatica, Vol. 1, No. 1, Sept. 1970, pp. 129. Pai, M. A. and Rai, V., "LyapunovPopov Stability Analysis of Synchronous Machine with Flux Decay and Voltage Regulator", Int. J o u r n a l of Control, Vol. 20, No. 2 , Aug. 1974, pp. 203212.
31.
CHAPTER V
5.1
Introduction
The computation of tcr for a given fault using Lyapunov's method involves three distinct steps, (i) Constructing a V(x) (ii) state defined by V(x) for the final postfault stable system. Integration of the faulted system Constructing a region of attraction around the postfault equilibrium < V Q r  (iii)
equations until V(x) = V which then yields t J cr While step however, (i) step
cr (iii) is
It is,
practical application of Lyapunov's method ever since the method was first proposed in 1966 for computinq tflr directly instead of by repetitive simulation t [1], For the past decade and a half, considerable research In general, it was found that although In view of this effort has been devoted to this problem.
r
inconsistency in results, there was not much enthusiasm on the part of the power industry to make use of this technique either as an aid for simulation or for transient security assessment purposes. Fortunately, the picture is quite optimistic now due to some recent breakthroughs on the research front [2,3,4]. Tn this chapter:, we shall discuss these new techniques in detail while briefly alluding to earlier efforts which gave conservative results. However, the formalism of the earlier methods is in
137
that the new approaches were possible precisely because of a deeper insight into the concept of region of attraction via the notion of invariant sets 5.2 Methods for Computing Stability Regions in Power Systems
Theorem 2.7.5 on the region of asymptotic stability, which is pertinent to our discussion, Theorem 5.1. is repeated below in a slightly modified form [61. Consider the autonomous system = 0 Suppose that the region R = (xV(x) (5.1) < k}
x = F(x) , F(0)
R converge to the origin as t * J. The assumption on negative definiteness of V(x) in R can be relaxed to the condition: of 5.2.1 (5.1) V(x) < 0 in R and does not vanish identically along any solution in R except the null solut ion.
(i) A practical method of estimating k was proposed as follows Let V(x) Let Vcr be the lowest value of V(x) on the surface V(x) = 0. region R  :xv(x) and is contained in the exact domain of attraction. bus system in Chapter IV The method, however, problem. (Sec. 4.5)
be positive definite and V{x) be negative definite near the origin. Then the bounded Zubov's method < V J) about the origin is a region of asymptotic stability
discussed in Chapter II (Sec. 2.14) and applied to a singlemachine infinite precisely uses this formulation. is unattractive from a computational point of view for
(ii) A second method is to generalize the results of Walker and McLamroch [7], as applied to the single machine case in Sec. 4.0, for the multinonlinear case and apply it to the multimachine system (8). In Sec. 4.7.2, it has been shown that the multivariablc Popov criterion
is applicable only to systems containing negligible transfer conductances, since only then that the nonlincaritics satisfy a sector condition around the origin. The state model for nonuniform damping case and with zero transfer (4.58) as (5.2) g = Cx The nonlinearity ^(o) ^fc*0^ satisfies the sector condition a. f k ^k^ in the region conductances is given by x = Ax  iif (o)
* k=l,2,...,m (m  n(nl)/2)
138
The Lyapunov function using the multivariable Popov theorem has been shown to be of the form V(x) T = x Px + q4i.(o^)do
m
2 1=1
T ie
ck 
<
k~l,2,...,m
(5.5)
l k2
A oneshot minimization of V(x) on the boundary of this polytope will again result in a large number of nonlinear algebraic equations which can be computationally burdensome hyperplanes 18]. [15]. An alternative is to consider one over all such hyperplane at a time and then choose the minimum of V(x)
Even then the computational extorts arc not reduced .substantially. However, the mathematical formulation underlying this approach leads to an which is related to the physical aspects It is also related namely, the modes of instability. approximate method of computing V _ of the problem, next. (iii) It is well known that surrounding the origin (posLfault SEP)
to the method using the unstable equilibrium points which is briefly described
there are a large number of unstable cguilibrium points satisfying the postfault equilibrium equations 0 = Ax  Bf(Cx) (5.6) i.e. 0 = F(x) (5.6) in terms of the physical variables to get a better
In terms of S^'s and co^'s, the model corresponding to (3.17) as u>. = (D./M )W + p  P i=l,2,...,n
(5.2)
is given by
*in * *.n
i=l,2,...,nl
0
and 5.
= 0 in
(5.7).
This implies
Y.
P . 1=1 = P . mi 
139
E.I
ft..
^ 0
(5.8) it is
Since m^ S represent the speed deviations from the reference speed ui speed without affecting equations in (5.7a), we
*
easily shown that the condition (5.8) can be met by changing the reference (5.7). If we add all the equations
on t + .
the postfault system are obtained from the equations ^ Although =0 ; (5.9) PA  P e.  0 i=l,2 (n1) n (3.9) variables only
(angle differences)
V>L are
P e = 0
11,2,...
(5.10)
In the case of uniform or zero damping, it may bo shown that the equations
U>j  0
(5.10)
or
(5.7)), there are a large number of other solutions of (5.10) or which represent unstable equilibrium points
(u.c.p.'s). The exact number of such u.e.p.'s surrounding the origin has been a matter of much speculation and discussion [912J. 0 and v(x) the polytope. However, it is not relevant to our discussion regardinq the Inside the polytope (55> we have V(x) > < 0 and also origin is the only equilibrium point inside The u.e.p.'s therefore must either lie on the boundary This conclusion is also consistent with (VV) T F(x) = 0 at each of the region of stability.
the fact that Lyapunov's method yields only sufficient conditions. Since F(x) = 0 at the u.e.p.'s, V(x) u.e.p.'s. Hence, V(x) can be computed at these u.e.p.'s and the minimum or these values can be designated as V c x third method of estimating the region was precisely the method suggested i n Ref.
\1]
of attraction.
in 1966.
However,
it was quickly realized that Two of the most widely or (5.11) were the
computation of these u.e.p.'s was not an easy task. used methods to compute the u.e.p.'s from (5.10) DavidonFletcherPowcll [10,11,12]. (DFP) Method
140
has to be computed at the large number of u.e.p.'s makes again computationally unattractive. However, throuqh
a combination of mathematical and physical reasoning, it can be shown that the u.e.p.'s of interest lie in the proximity of certain points on the boundary of the polytope defined by (5.5), and called as approximate u.e.p.'s. These points are easily identifiable and can be associated with the physical of loss of synchronism. of possible u.e.p.'s as 2(2nil) . 5.2.2 system Approximate u.e.p.'s and Modes of Instability (5.2) and a typical hyperplane defined by
(5 12)
modes
k  k* " *ki where cR = (ck] , ,c^ 2nl* is of V(x) on this hyperplane. that (5.9)
thc
row of
We are
seeking
minimum
is tangential to V(x)
at that point.
The conditions
v(
Where
Xk is a constant.
Assuming V(x) is to be of the form (5.4) and after some algebraic manipulation, (5.13) becomes 2Px + 2CTQf(g) where >. = in in o. (5.14),  CTX  0 (5.14) With a = Cx substituted
(0 ,...,>.,,.., 0)T, Q Diag (q. ) . (5.12) and (5.14) and the 2nl unknowns x.
together constitute a set 2n equations Let thc solutions be. denoted by x and 18] (5.15)
Since there are 2m hyperplanes corresponding to over all these hyperplanes gives an estimate
of V
The preceding theory is general enough for any physical system described by In fact, only it may be verified that not all ^k* * (n1) of the m ^ (n1.) lisiearly independent. ) o. 's.
K
a linear combination of the independent and the rest combii partiti oned as
(mn+1)
Si
0 =
141
where the subscripts J and I) refer to linearly independent and dependent o 's.
fv
i 1 n1,n
The
o, k
it
kn
hyperplanes and it
is
The case of single nonlinearity can now be considered as a particular case of the above formulation when o = c, a scalar. Then  Min(?,2 ,1 ^1) and <*26s) f(u)do (5.17 ) = ,s (vi23 )
IcP
c ]
142
In the case of power systems, Lhe boundary of the polyLope defined by surface [5] on which the term corresponding Lo the integral (5.4) has a relative maximum
(5.16) in the
in Lhe angle space corresponds approximately to the potential energy boundary Vfunction in (potential energy term). Lerm
The quadratic Lerm representing Lhe kinetic energy has a relative minimum and is also much smaller in absolute value compared to the Integral near the tangency point of V(x) (5.14) becomes C' (Of to) Since C = (0  A) 0 n n1 (5.18 ) to the hyperplane. the quadratic term in comparison to the integral term in V(x). Thus, we can neglect Equation
S), Eq.
(g)   A) (5.19 ) Equation (5.19) is a set of (n1) equations. (mn+1) (n1) Since only (n1) of
the o'^'s are linearly independent, the other as a linear combination of the and A . Eq. A solution to
J.
J.
(5.13) which satisfies the restrictions on other o^'s (i?00 is the The minimization can also be performed on the intersection
tangency point.
of more than one hyperplane at a time. The structure of Eq. system. (5.19) will be explained for the case of a 4machine (3.47) I (5.20 ) as
C is given from
C =
0 0 1 1 1 0 0 1
Choose the linearly independent set of a^'s 0 as :; 14' . s 6,  <5^ This implies choosing machine 4 as a reference machine. We have = i a5 = ffi ~ 3'
OA ~~ & ~ . A r 
1
"14
c^,
a^,
is
'J'nen
(5.19)
0
becomes
<Wic3> *
143
" (34f4(0l"fc21)
+ g f U
6 6 2l"03}
V2
<5'21
q3f3(o3}  ^5 ^5 ( 0 ^0 3 )  q6f6U21o3) = 0 The above equations are to be solved for tf^r $' and that and 0 ^ satisfy sl^., < c^ < such
tgg
< c
31*
^S^se
are nonlinear equations and can be solved by any of the well known methods for solving a system of nonlinear algebraic equations. However, one needs Thus, for the good initial guesses (starting points) and these arc obtained by associating physical modes of instability with points on thc polytope.
Fig. 5.1.
losing synchronism by acceleration with respect to the reference machine while other machines are at their postfault SEP's. This initial guess is also now an approximate unstable (a.u.e.p.). i.e. c2 = i*e" In the polytopc of Fig. and = i** ~ ii  ^24' = tfg = 0.
tnen t i
5.1/ If we had
point corresponds to machine 2 decelerating wiLh respect to machine 4 while other machines are stable. and thus generate Similarly, we can take the other hyperplanes in all 6 approximate u.c.p.'s numbered 16 corresponding
to one machine losing synchronism either by acceleration or deceleration. To this we must add two other a.u.e.p.'s on the polytope, one corresponding to all the three machines accelerating with respect to the reference, i.e.
:7
1 ~ *11' 2 = ^21' 3 ~ *31 **>oint 7 in Fi9 51) and the other corresponding to all the three machines decelerating with respect to reference machine, i.e. ~^^2' 2
=
~*'22'
144
3  ^32 (Pint
8 in Fi
9 $.1).
these Lwo points correspond electrically to machine 4 losing synchronism by itself. In all, we have eight physical modes For an
n 1 n1
of instability of one machine losing synchronism. nmachine case, these will be 2( ~ C1+ Cn_1) = 2n.
We next take two machines going out of step at a time either by accelerating or decelerating together. Suppose we choose the hyperplanes 02
= l21  1  2S4

hi
 * "
25
34
Then following the same formalism we will get three equations but the unknowns will be Physically, and the two constants >2 and this corresponds to machines 2 and 3 accelerating c3 * 
^34*
~ 0.
This is
point: 9 on the polytope in Fig. 5.1. together, i.e. together type. (a) 2(n C^).
total of six possibilities of two machines accelerating or decelerating These are points 914 in Fig. already have considered three machines accelerating or decelerating (points 7 and 8). Thus, in a 4machine case, we have in all 14 These can be summarized into the following types: Each machine going out of step with respect to the reference machine by acceleration (7r2^4,0,0> , (0,n26^4,0) , (0,0,7126^} (5.22a ) (b> Two machines goinq out of step at a time by accelerating
( T r 25^,
TT
26^4, 0),
(n264, 0,
n26^), (5.22b)
(0, n2624 , "26^) (c) All the three machines going out of step at a time by accelerating (n26^4, (d) Repeat (a), acceleration.
71
(5.22c)
(b) ,
of synchronism by deceleration instead of (5.22d ) In the case of n machines, we can generalize the result as follows: arc *
n
There
145
(5.23 )
k terms
(nIk)
Thus, the total number of modes of instabilities and hence the corresponding
"i
n_1
C,
 2(niC.+n1C:)
...n3Cn .)
 2(2n1l)
[93.
a large number of Uve.p.'s is not practical and hence there have been efforts to limit the search to a credible number of u.c.p.'s. next. 5.3 Quick Determination of V Through Approximate u.e.p.'s
T77T3T 
cr


ffi


It has been the practical experience that generally whenever the system loses synchronism one of the machines will pull out first either by accelerating or decelerating. Hence, it was proposed that the search for u.e.p.'s be limited to those in the proximity ot the following approximate u.e.p.'s: (ij (ii) (iii) (n1) of the type (0,0 ,.. . n26^, (n1) of the type one of the type
0.
. . 0)
i=l,
2,
. . . ,nl.
it.
x, ii
(iv)
one of the type (iii) with +IT replaced by n. there are 2n approximate u.c.p.'s. [11] It has been shown by in the neighborhood Thus, evaluating V(x)
Thus in all
19J
of an unstable equilibrium point are very small. of V(x) be 1. 2. Compute V(x) at the 2n approximate u.c.p.'s. at the neighboring u.e.p.'s.
Choose among the 2n approximate u.e.p.'s the one Which gives minimum value of V(x). desiqnated as xu Let this u.e.p. be
3.
appr Using NcwtonRaphson or DavidonFletcherPowell method, compute x with x" as the starting point from (5.10) appr or (5.11).
146
This method has given satisfactory results on relatively small sized systems. An improvement over this method is to limit the search to n 114J (fc = l , 2 , .. . ,n) where P  LPX points as follows: (i)
~ .....
n_1)
of conservatism in the computation of t.^ appears as the size of the system The reasons for this as well as methods which overcome this basic deficiency are discussed in thc rest of this chapter. 5.4 A
V
lacuna
in
the
method
based
on
evaluating
V^r
as
the
min
lating the mode of instability based on postfault configuration only. This is incorrect as thc following reasoninq shows. assume that the postfault configuration is the same for the
the different
facult locations, i.e. there is no line switching. The use of same value of V . implies that irrespective of the fault location, mm the same machine/group of machines goes unstable corresponding to the u.e.p. which determined
v m^n
Physically, this is not true and we know that the In general, the critically cleared
actual mode of instability is influenced by the fault location, the type of fault and the fault clearing time. but unstable trajectories for different fault locations will exhibit different modes of instability and hent;e will pass through or in the vicinity of different u.e.p.'s. size increases. This is especially true as the system faults Any In small systems with a tight interconnection,
at different locations may result in the same machine going unstable. new approach to the problem must therefore take into account the faulted trajectory and the mode of instability it induces in the postfault system. controlling u.e.p. relevant u.e.p. V cr for the particular fault. Hence, computation of V becomes fault dependent and for each fault there is a relevant or a should be thc value of V at this
to the problem and we shall discuss all three of them in varying detail. All of them have a fast method of computing the faulted trajectory through some approximation. The methods are: (i) Method based on acceleration criLcria and determining the controlling Method based on computing
the controlling u.e.p. through a l)av idonFletchcrPowcll algorithm after getting a good approximation for it using the faulted trajectory and principle singular surface properties notion of a potential encrqy 13,5J. (iii) Method based on thc
147
boundary surface surrounding the postfault SEP and use of the faulted trajectory to compute V"cr directly 12,5]. 5.4.1 In Ref. Approximate u.e.p. and Acceleration Criteria [4]
[11 which presented Lyapunov's method tor stability analysis for it was observed that the machine having the largest ratio
P
of initial acceleration to inertia constant at t accelerate faster than the other machines. machine would become unstable first. be chosen would be (0, . . . ,*2&^n,  0) Hence,
where k is the machine having the largest ratio of acceleration to inertia. This conjecture may work well with small systems but is likely to give inconsistent results in larger systems. In Ret. [4] the accelerations are computed not t  0 1 , but A little ahead on the faulted trajectory by which time the pattern of instability will have been established. This method incorporates the procedure outlined in Sec. as the first conservative estimate of t the likely mode of instability and the t_^. 1. Compute the value V cr fappr the and then updates both cr The steps to be followed are: 5.3
2.
Execute forward numerical integration of the faulted system. At each time step compare V(x) with V Let t . correspond to the case when V(x) = V cL * cr,appr cr,appr Let k be
3
the machine with the largest acceleration at t  tcL'I'h i s machine may or may not be the same as in step 1. Corresponding to machine k, find the value of V(x) at Lhe a.u.e.p. (0,. ..
,7
2^n, ,0).
If
necessary, the exact u.c.p. xu can be computed using Lhe a.u.e.p. as the starting point and Vcr can be corrected. 4. Continue numerical integration until V(x) = a qood estimate of actual L^. What the above method needs is a fast numerical method to integrate the faulted equations from t  0 to t = tcLprefault steady state soluti on [14]. (t t
6 v cr
In fact,
numerical integration
6.(0
= *.[
P)
it )
il,2,...,n (5.24)
t t
is the initial instant when the disturbance occurs, o denotes the time when the system reaches its next J
148
fi.Ct ) (5.25)
(P) E
,A
f t )
( t
V o>
c
p
t"
11*2 ..... n
1=1,2,...,n
(5.26)
i at
=
mi "
(P
Gi<S>
i = 1
'
2 n
(5 27>
'i2''^)
S7
miPci((0+)>
i=1 2
'"
8 p,
i o   *ir2}'<>
i=i 2
' ................
(s 29)
For p " 4, Pe^() therefore has to be differentiated twice to obtain rt!3)(t+) and
$W
5.4.2
The above method has been tested on a large number of test systems such as 3, 7, 9, 15 and 40 machine system with satisfactory results. The results for the 7 machine CIGRE SyBtem is presented here. data Tor the system. Figure 5.2 shows
the single line diagram of the seven machine system and Table 5.1 gives the The values of tQr obtained by this method are
v at m^n
the In
u.e.p.'s) as well as thc stepbystep method and are shown in Table 5.2. Column 1 of the table indicates thc bus where the fault is applied. column 2, ] prefault configuration. because of line switching. indicates that thc posttault configuration is the same as D indicates that the two configurations differ It is observed that while the standard
GENERATORS Bus 1 2 3 4 5 .6 7 BUS 2 4 6 7 p base (MVA) 100 100 100 100 100 100 100 P (MW) 200 650 80 90 X (*) i 1 ) 7.4 11.8 6.2 4.9 7.4 7.1 8.7 0 (MVar) 120 405 30 40 LINES Bus 1 1 2 2 3 3 4 4 4 4 0 7 R (ohm) 5 5 22.8 8.3 6 5.8 2 3.8 24.7 8.3 9.5 6 24.7 X (ohm) 24.5 24.5 62.6 32.3 39.5 28 10 10 97 33 31.8 39.5 97 wC/2 (us) 200 100 200 300 300 200 200 1200 200 300 200 300 200 M (MW s2/rad) 6.02 4.11 7.59 9.54 6.02 6.77 5.68 LOADS Bus 8 9 10 P m (MW) 217 120 256 300 230 160 174 E (P U ) 1.106 1.156 1.098 1.110 1.118 1.039 1.054 P (MW) 1.00 230 90 0
fi 
3 4 3 10 4 9 5 6 9 10 8 8 9
( ) These values include the transformer's reactances and are expressed on a 100 MVA base. Standard Acceleration StepbyStep _ o Lyapunov Fault Final Criteria Table 5.1 Conf Data of the CICJRE 7machinc system Location jguration Method Method CJ (Reproduced from tRef. [14]) 225 kV 2 tcrU0s> tcr(10"2s) conservative estimates, the present method gives excellent agreement 1 I 3536 2627 3536 with cstima1 for fault at bus3637 fl2 2 actual t^r except I for a conservative 4142 3435 1 2528 3940 and3a pessimistic estimate at bus #8. 3940 4 1 5051 3031 4748 5 I 3536 2627 3536 6 I 5253 3940 5051 8 D 4445 4041 5051 6 D 5152 3435 5051
(io
Table 5.2 Critical clearing tim e by acceleration criterion and standard Lyapunov method for the 7machine system Fig. 5.2. 7 machine CICRE test
COMPUTATION OF
STABILITY
MSG!ONS
150
The
computing time by this Method compared to the stepbystep method is improved by a factor of 100 141.
(with
reason for conservativeness of Lyapunov's method was systematically [5J and they proposed a methodology which takes into
account the dynamics of the faulted system and also the transfer conductance The concept of a controlling u.e.p. was introduced and discussed Thc notion of a controlling u.e.p. for first swing 12J. While thc former uses the center differ in the way the transfer [bl provides a proper extensively in the context of both first swing instability and multiswing instabilities. instability is also contained in Ref. frame. The two methods, however,
of angle reference formulation, thc latter uses the machine angle reference conductance terms are accounted for. rationale for taking invariant sets Reference
into account thc faulted dynamics via the concept of of kinetic and potential
energies and make use of the properties of the potential energy function. Invariant set and region of attraction Consider a set of disturbances configuration. clarity, and (faults) i^l,2,..., at different
locations which, after being cleared, will result in the same postfault Typically these will be faults at the yencrator/load buses For conceptual (3.34) which are selfclearing without any line switching. (3.15), (IMuHod phase) x Pj(x,P*), x(0) xn, 0 < t < where I*, thc
Phase A
is the parameter vector different for each i1n fault. Typically, elements of this parameter vector equation. Please B (Postfault phase)
are
x = F2(x,P^r)
The solutions of the equation 0 = 2 (x,Ppf") are the equilibrium points of the postfault system in the state space. Let xe be the SEP and the remaining equilibrium points will be the u. e. p. ' s. Basically, the transition from the state Xq to x& is what is of interest to the power engineer. The two sets of d.e's  xQ 0 < t < tc (5.30) x  F2(x,P ), x(tc) = xt c form a discontinuous set of d.e*s describing the motion of the state variables. Given a certain t > if the state of the The maximum Because of the system qoes from x q to xe, the system is stable. t for which this is possible is called t . q cr
pf
x  F^X/P*), x(0)
t  tc
151
inherent mathematical difficulties associated with stability investigation of a set of discontinuous d.e's via Lyapunov's method, we consider only the postfault d.e and construct a region of stability around x^. Def i ni tion An invariant set for a system of equations x at t tQ in fl belongs to ft for all t > t^.
f(x), x(0)
 xo is defined as
any subset of the state space having the property that a trajectory starting For example, the equilibrium state xg is an invariant set and so is any trajectory of x
f(x).
In the power system problem, we define the invariant set for the postfault system as the set of all trajectories of the postfault system x = F2 (x,Ppf) , t'> t.6 lie on the faulted trajectory =
x 0
(5.31)
* = Li^'EJ <
for the fault i. Tn Kiy.
(5.32)
5.3, the faulted trajectory is shown by the thick line arid the
invariant set by thc dotted trajectories. Thc region of attraction is defined for thc postfault SEP, *c with respect to this invariant set. Theorem 5.1 Let ft The tollowinq theorem is directly applicable.
be
FgCXfE^')*
>
to ii. Let V(xx^) be defined on G with v{0)  0. Let s(k> denote the set s(k) { x L
(i) V(xx^) is positive definite and decrescent on (^0)/ and (ii) is negative definite on s(k^) then
m
(1<0> is contained in the region of attraction of xfi. now interpret this theorem for thc power system Along the faulted trajectory for fault V(xx ) e
we shall problem.
computed for the postfault system increases with increasing t. For to t < tcr/ the trajectories of the postfault system converge is For and for this subset of the invariant set V(xx^)
positive definite and V(xx^) is negative definite. hence V (xx^) is not negative definite.
t > tcr, thc postfault trajectories diverye away from xc and Consequently, for this subset of the invariant set there does not exist a vol id Lyapunov function. The value kQ is the value of v(x) at t = t on the faulted trajectory. For t t , the postex cr
152
fault trajectory becomes unstable and theoretically passes through but practically goes fault i. In the vicinity of x
1
lor
This is analogous to thc Single machine case where thc critical The critical
trajectory is shown hatched in Fig. 5.4. For another fault u f j location P.., we have a different controlling u.e.p. x J (also shown hatched) . The computation of kQ  ^cr
eac 1
' fault
is
Fin. 5.3. Illustration of the faulted and pusL fault trajectories for various clearing rimes corresponding Lo stable and unstable vases.
Fig. 5.4, Invariant sets and critical trajectories for different faults. therefore the crux of the problem even though the postfault configuration is identical. If the system is conservative, V (x) remains constant at after the fault is cleared and hence the value of k at t = t is same as the value of V(x)
153
x = x
Ui
where x
Ui
cr
in Lyapunov; stability analysis of power systems until considered the value of k u i each of the x postfault configuration.
irrespective of fault location for the same It is therefore clear why the
results obtained by such a procedure were conservative in terms of t . Since k has to be computed for each fault cr o condition, the algorithm must emphasize computational speed. Methods and (iii) cited at the beginning of Sec. be discussed. 5.4.4 Method Based on Computing Controlling u.e.p. 3]" Through DFP Routine (ii) 5.4 which achieve this will now
The center of angle formulation is used. damping) M .oi. = P. i i !' = The Lyapunov function is given by V (fi,<T>) I = 5 1 i~i n
T. T.
n 1 1
P .
i ei
; P, My
k COA f
(6)
l 
(5.33)
i = l ,2, ,n
M . LU T ii
c,
1
P. (6.67)
n C. . (cose . .cosO? .)
i=l j=i4l
*V
(5.34)
energy function)
Total change in roLor kinetic energy relative to COA Total change in rotor kinetic energy minus the change in COA kinetic energy (ii) n >: i1 P. (O
1 1
.of ) 1
Z i=l
n P. (6.6?)
1 1 1
J:
Change in rotor potential cnerqy relative to COA = Chanye in rotor potential encrqy minus the change (iii)
C.j(coseijco a ?
.)
(iv)
(5.38)
Change in dissipated energy in branch ij The kinetic energy term (5.35) is denoted by V (<o) , the sum of
154
(5.36) (5.38)
and
(5.37)
is denoted by V ^(0)
and
denoted by V^(fl)
To successfully compute thc u.e.p.'s in the presence of transfer conductances, the center of angle (COA) 0 ^  0 in (5.33). This gives formulation is advantageous. The u.e.p.'s are the solutions of the equations obtained by setting f&v =
k<M
 0 Since P C l
i 
ci 
COA
(5>39)
i1,2,...,n
and P,,,,. are functions of differences of 6 . . ' s only COA I j n and 0 1/M Z M.6., it is sufficient to consider only the n 1 1 n i1 first (n1) equations in (5.39) namely
V2> =
i 
"ei  If
(SEP) 6s.
>cm
i = 1
t 5
401
Equation
(5,40}
has several solutions, one of them being the postfault Surrounding this are the unstable (5.40). One is At all the equilibrium points (stable or unstable)
the NewtonRaphson method Which is discussed adequately in the literature The second method is by minimization of a scalar objection 2 I f(0) which is physically thc total bus power 1 i1 " M. mismatch. The presence of the term P in each equation ensures that the transmission power loss is not absorbed by thc reference The While NewtonHaphaon method
machine alone but by all machines in proportion to their inertias. minimization is done ucinq the DFP technique.
is suited well for computing the stable equilibrium point, it lias convergence problems with unstable equilibrium points and hence the DFP technique is preferred The u.e.p. condition. [5. to be computed is the controlling u.e.p. The procedure consists of two parts. for the given fault
(i) Calculate thc fault trajectory of the system, i.e. integrate Kq. (5.33) with parameters corresponding to the faulted network and initial conditions at t = 0. (ii) Kxon the faulted trajectory, estimate the likely mode of instability and hence the u.e.p. We will elaborate upon each of these aspects. Calculation of faulted trajectory: Since the purpose of this computation a very is only to know in which direction the faulted trajectory is moving,
155
Hence, we approximate
+ fl.cosnt
(i=i,2,...,n)
(5.41)
(i^ and n are computed on the basis of two power flows. Thc first at The second, along an is used to compute r\. The
the instant of fault application fixes vectors u and 6 tor a given frequency
approximate trajectory shortly after the fault, details arc contained in Ref. [51,
simplification is based on the results obtained by the approximate method and the exact simulation for the duration of the faulted period. already seer, in Sec. trajectory through a Taylor series expansion. Estimating the controlling u.e.p.: sum of
v p()
5.4.1 another way of approximating the faulted In the Lyapunov function (5.34), the
t.(0)d0.
(5.42)
In the
local region around 6" in the angle space and referred to as the potential energy function is convex. The principle singular surface is The closed boundary of this region is denoted as the principle singular surface [5,16]. the multidimensional analog of the region of dynamic stability which in a 2machine system corresponds to the peak of the power transfer curve. In the ndimcnsional case, if we define 9in  (J.  Cn (i=l, 2, . . . ,nl) as the relative rotor angles, then a polytope defined by Gj_nl 1 ^/2, i=l,2,... ,nl and l^i'jl 1 A2 i,1=1 ,2,...,nl can be inscribed within the principle singular surface [IB]. The faulted trajectory is integrated until it intersects the principle singular surface. To know this point of intersection, the quantity n 2 F(ft) = 7. f.(0) for the postfault system is monitored and when F(0) reaches a maximum, the angles 0SS are close to the boundary of the principle singular surface [5], This conclusion car. be justified by the (5.42), in the absence of transfer
t 1
following supplementary reasoning. From conductances, the gradient of ^p( ' ) accelerating powers t^[0).
Euclidean norm squared of the gradient of the potential energy function. Hence on the boundary of the principle region, F(0) is a maximum. Interestingly, if the Jacobian of f(6) is computed and its
on the principle region boundary surface one of the eigenvalues becomes zero [18). Step 1 For the given fault, determine thc fault trajectory approximation parameters, The various steps in the algorithm are summarized as follows [5]:
a^f
Step 2
il,2,...,n and n.
Calculate the parameters of the postfault matrix Y , and s compute U from (5.40) using the NewtonRaphson or DFP technique with the prefault equilibrium point as the initial guess. Step 3 Usj ng approximation of Step 1, power mismatch function K(6) The angles at this time are 6 point of the Step J, Starting with 0"s, a vector 6SS  6s is formed and normalized to form thc direction vector h. Step 5 Solve the one dimensional minimization problem min F(0(z)) F(M (z*>) where Step 6 With 3 as the starting point, solve (5.40) using the DFP minimization technique to obtain the unstable equilibrium point 0U. fault. The following comments serve to clarify Steps (l)(6). arc selfexplanatory. signifies that maximum synch ranizing capability of the postfault system is reached. established. The basic idea in Step (4) Also by this Steps (1) and (2) is maximum V(0,0U) gives V^r for the given v)(z) 0SS + zh and z > 0 U(z*)
ftR
find the time at which the (postfault) (very close to) the intersection
k GU
time, effects of the faulted trajectory will have been is to construct a s corresponds direction vector which points in the direction of the u.e.p. corresponding to the fault case under study. to the origin in the polytopc defined by {5.5) and among the different choices possible for the direction vector, this direction is chosen from the point of view of computational simplicity and determining the first swing instability mode. Tn Step (5), the minimization of the power mismatch function along the direction vector, constructed in the previous step, defines U which is the point on the vector closest to the v u controlling u.e.p. u.e.p.
U
In Step
(6)
U
v\j(9u)
is an
and hence the following approximate formula is used to This is derived from the integral expression s by assuming a linear path of integration from 6 to
n1 V (<."> =
au + o"  of Z  T.
i%
i 1
~i  2
Wmsg 5.4.5
ine.iD. . 5.5.
A numerical example illustrating the method is presented in Sec. Potential Energy Boundary Surface {PEBS) Method
in the previous section, we have seen that for a particular fault condition, an accurate estimate of V~cr can be obtained by computing V(x) at the controlling u.e.p. xU. u.e.p. ^ 0, Since at an
u
Vp]i.(l! ) = Vp(9 )
+ Vd<0 )
U
i.e. 6 in the
the
region from the state space onto the angular cubspace of and U consider Vp(6) + V"d(G) i
PE12>
l*>
p constant)
i.e.
space
can
be
projected on to the C space and interpreted as the VpE(0) = constant curves. These are known as thc equipotential curves. These curves arc closed in a small region around thc stable equilibrium point the general shape shown Surrounding called the saddle points, F:ig. (Tj^ and U,,) 5.5 the SEP are a number of u.e.p.'s Qu. (SEP B s but otherwise have Cor a 3mach i ne case. At some of the u.e.p.'s i*e. (5.13)
E = = ________ E = o.
The cquipotcntial curves around the saddle points arc not closed curves. There are other u.e.p.'s the SEP, (U^) in addition to the saddle points around which Another way to characterize is through thc At the SEP, the the equipotential contours are closed curves.
eigenvalues are negative, at the saddle point there are both positive and negative eigenvalues and at thc other u.e.p.'s the eigenvalues are positive [18]. origin). maximum. In terms of the potential energy, the F>F.P 6s has the minimum At the saddle point type u.e.p.'s, the energies have a relative We consider the u.e.p.'s in an internal of 11 to +11 around O . potential energy (zero if the postfault SEP has been transferred to the minimax pattern and at the other u.e.p.'s the energies have a relative We can also interpret
thc equipotential curves in the relative rotor angle reference .0 n frame fljn as in Fig. (5.6). S is the stable SEP while I?2 and t!^ v.r
.t
/ .::.bJ <.
; i ::
are the saddle point type u.e.p.'s and energy is a relative maximum. in Figs. (.5.5) and (5.6)
th
The motion of the system trajectory can be projected on to the angular space and the loss of synchronism context of thc u.e.p.'s as (instability) phenomena can be viewed in the
159
20 0
Fig, t>.b> Equipotent 1 al energy surfaces (solid lines) and the potential enprgy boundary surface (dotted line) for threemachine system {reproduced from Ref. [5J).
follows
(Figs.
r f a fault
the vicinity of one of the saddle point type u.c.p.'s. unstable trajectory will in general
different location of the fault, the critically cleared but pass in the vicinity of faults. Such a another saddle point type u.e.p. even though the postfault
contiguration may be the same for all the is termed as the controlling u.e.p,
saddle point where the critically cleared trajectory escapes flu for the given fault. It need not necessarily be the u.e.p. with the lowest energy as was assumed in ail work prior to 1978 and which resulted in conservative estimates for t . cr For t > t the trajectory cr u
will cross the PF.BS at a point close to the relevant u.e.p. 6 if the system is first swing unstable. For multiswing unstable cases, the unstable trajectory may cross the PEBS at an altogether different point on the PF.BS. If the fault is cleared before tcr then the trajectory will oscillate within the PEBS and ultimately settle down at the SEP 6 if there is damping in the system. If we restrict Lo the case of first swing unstable cases, then unstable trajectories includinq the sustained fault trajectory cross the PKBS at a point such that the following hold true (Fig. 5.7). n . (i) The mismatch F(6) = Z f.(0) is close to zero indicating 1 i = l that the unstable trajectory is crossing the PEBS at a point close to the controlling u.e.p. (ii) Vpg(*J) is maximum and (m) is minimum at the PEBS crossing.
It is therefore logical to assume as a good estimate of critical energy the value of VpE(0)  Vp(8) * Vj(6) at the PEBS crossing for the sustained trajectory. multiswing cases with satisfactory results (51. this procedure for practical PEBS cross i rig. purposes, Experimental results In order to translate
indicate that the same procedure can be used for estimating Vcr even for we need a criterion to deted the
161
The
=> V. (w)
K 
+ V p
(0)
{':>. A
V (i7>,0)

V, (ffi) + V (6) K p i.
(H.5.3.
^ 3 6,
6i)
(5.45)
ary
= (T + ^) i (5.46! Since the transfer conductances arc negligible, the postfault system is conservative and V(t5,0)  0. Hence for nonzero to, we must have
I 
 if
<547:
Therefore, the direction of the torque vector T is in a direction negative to the gradient of V^(), i.e. T = VV^. By definition Z p *s ^n Hence,
a
of increasing V  constant contours in the angle space. decreasing Vp = constant contours. Consider a trajectory crossing the PEBS.
PEBS and containing the SEP, the projection of T on the velocity vector is in a direction opposing the motion, i.e. towards the SEP. outside the PEBS, i.hc projection of T is in the direction of the velocity vector. This suggests that the dot product of VV^ with the velocity vector chanqes sign in the vicinity of the PEBS. orthogonal to PEBS. it In Fig. (5.8)a
trajectory is shown crossing the PEBS such that the velocity vector is is easily verified that inside the PEBS the dot When the velocity vector the point where Tuj since product Toj < 0 and outside the PEBS, > 0 and on the PEBS, Tu>  0. does not cross the PEBS orthoqonally, does not lie on the PEBS. However,
0 In
changes sign
rapidly near PEBS, the change in sign occurs very close to the PEBS. the case of a conservative system, n
163
>
+ V.  0 for a conservative system, we have V. P " inside PEBS VR < 0 and outside the PEBS > 0.
these relations imply is that inside the stable region potential energy increases and kinetic energy decreases. Decrease in V. means decrease in The
jic. I
and prevents the system from getting outside the PF.BS. Consequent\y, when V or V. changes sign can be p K
interpreted to be the instant of time when the trajectory crosses the PEBS [2]. In the presence of transfer conductances, the above observations hold good only approximately. is sign indefinite. [5]. The reason is that although v<u,0) The deviation of V( uj,0)
>
0,
V(co,Q)
significant and if some damping is present V will be negative semidefinite There are two methods for monitoring the PEBS crossing for an unstable trajectory. (i) Since V(u,o) (ji) + V (0) t Vd(G) and since tf^lS) is the
same whether the transfer conductances are present or not, the instant of time when (ii) = 0 is taken as the PEBS crossing [2).
Criterion
%
\
Comment zero transfer conductance case zero transfer conductance case Both zero and nonzero transfer conductance cases
Table 5.3
In Tabic 5.3, the criterion tor crossing the PEBS based on the signs ot V and V. P kpiiBS 5.3)
K
arc summarized.
on the sustained fault trajectory and for that particular fault, 1 p PEBS cr Application of this criterion leads to difficulties when criteria can be used to compute V {tp^gg) (2J.
is an estimate of Application of this criterion transfer However, another characterization of PEBS to be discussed
conductances are present requiring a correction factor tor V^^ next is valid tor both zero and nonzero transfer conductance cases. Constructing the PEBS [51
PEBS
Fig. 5.8.
(21).
The nature ot PEBS can be understood through a 3machinc example with transfer conductances neglected. 0 ^ 1s arc linearly independent, V in the space of 6^ and (8) Since only two (5.5) is shown in Fiq.
the same as the prefault configuration so that there is only one SJEP whic h also is the point of mi nimu m potential energy. The two saddle points have energies of 3.2 and 7.2 respectively and the third u.e.p. has an energy of about 9.0. contours is the PEBS. The dotted line joining the u.e.p.'s and orthogonal to the equipotential The PEBS forms a "ridge" and can be starting defined by the following constructive procedure,
from the postfault SEP, go out in every direction in the angle * space. Along each direction fin d 0 where V attains a * P relative maximum. the PEBS. (i) System with zero transfer conductances Mathematically the PEBS can to zero. By is given by Vp(fiS+c*u>  Vp(9s) The set ot points 0 found in the manner characterizes
1 im  0
where u is the unit vector along which the derivative is desired. case
In ou r
W
_p
"
(Srif)
i
(5.49)
1 1
Sinc e on
the PEBS, V
[ 51
th
VfJ<i) + v (0) 1
y:
(5.5i>
1
fi

v^(8)
being
approximated as
^nnr 'i1 * dppr = _J:
(5.52)
dV *)fi  = 0
P
implies
l
n f
If 0
(5.52)
i.e.,
satisfies
n s
I i1
f, (0) (9af)
1 1 1
= 0
(5.53)
This is the characterization of the PEBS when transfer conductances are present. In vector form,
S
(0)(e9S)
= 0.
Denoting 9  e = 9, inside the PEBS, the quantity f and outside thc PEBS it is greater than zero. by analogy
wiLh l be zero transfer conductance case when f (0)0 reduces .w :iv  (  T ^ )  I ; i n I nee f (0) = . Within thc PEBS, when 0 is moving away from i)s, both VV and U are > 0 . Hence, When ft is moving inside the PEBS Note the towards 0s, bol.h VV^ and ft are < 0 .
iYp* '* *
product will
By a similar reasoning outside the PEBS the dot be < 0 and on thc PEBS equal to y.ero.
167
(5.48).
crossing. 15]
Potential
The energy function when transfer conductances are not neglected, contains an integral term which is path dependent, i.e. n1 D.COSO^dtO.+Q.) n
6 . i6 .
V'S) 1
i=i
n1
j=i+i n
e+0j
1
11
i=l ji+1
These terms represent the energy dissipated in the transfer conductances. The correct path is that of the actual system trajectory. (5.54) above can be
When the trajectory is known, the terms in Eg. evaluated using the trapezoidal rule as follows.
X
ij(m) '
ij(m"i)
lo i(m)
e .(m)
I^j(ml)
e.(ml)
and fi^(m), C^(ml) are the anqlcs at bus i and j at time steps is used in computing for the
respectively, and l _ j j ( n ) = 0 for all i,j. This potential energy approxim.it. ion faulton trajectory as well numerical
This
V^(0). Computation of Vgr, tcr by thc PEBS Method It has been pointed out already that u.e.p. and (ii) (i) the faulton trajectory
(sustained trajectory) crosses the PEBS at a point close to the controlling the potential energy V__(0) of the postfault system at Hence, this value can this point is approximately equal to the value of the Vfunction at the controlling u.e.p. for the first swing instability. be taken as a first approximation to V^^. computing; ^cr consists of the following steps. 1 Both thc PEBS and the iterative PEBS methods are strictly valid for the first swing instability case. However, fairly accurate results are also obtained for multiswing cases also. 2 Both thc PEBS and the iterative PEBS methods are strictly valid for the first swing instability case. However, fairly accurate results are also obtained for multiswing cases also. 3 Both thc PEBS and the iterative PEBS methods are strictly valid for the first swing instability case. However, fairly accurate results are also obtained for multiswing cases also. 4 Both thc PEBS and the iterative PEBS methods are strictly valid for the first swing instability case. However, fairly accurate results are also obtained for multiswing cases also. Therefore, the PEBS method of
168
(i) Compute the faulton trajectory cither by numerical integration or by one of the fast methods discussed in Sees. (5.4.1) (ii) Compute fT(0) (fi8s) both in the computation of V and Vp ( f i )
<e)
and
(5.4.4).
pE(6).
J
fgig) = V
(5.54).
V\,
Note that
parameters pertain to the postfault configuration. (iii) Inside PEBS, the expression T<fi) ( f l  R s ) = 0. is less than zero and (i> and (ii)
outside the PEBS it is greater than zero. are repeated until fT(9)(63s) value of Vp^,( )
H
Hence, steps
v cr
Iterative PEBS Method Following Ref. [2], a variation of the above method discussed below can At the end of step (iii), let thc value of Vfr Then, integrate the faulted equations and at the PEBS crossinq is thc updated ^ Monitor the postfault trajectory until . Repeat this procedure two or three times to Two comments are pertinent here.
v cr
improve the value of V c r be designated ris = clear the fault at V(x) value of V = v
the PEBS is reached. The value of VpE(0) improve the estimate of Vc.
187
(ii)
In Ref.
approximation for v^fti), the Lyapunov function due to Gudaru [19] is used. However, a correction to V(x) is then applied to obtain V since V(x) is sign indefinite. Also the PEBS is cr , fp g identified by the zero crossing of V. instead of f ( U ) ( * >  " ) .
JC
Numerical Example [5) A 10machine 39bus system known as the New England Test System is used to illustrate the computation of tcJ.. Fig. 5.0. references in the literature. The single line diagram is given in The line data and generator data are contained in many Two methods are used to compute the
critical energy ^cr and the critical clearing time tcj_. (i) (Sec. Controlling u.e.p. method 5.4.5). (Sec. 5.4.4). (ii) PEBS method
For both these methods, the results of tcr arc compared with the actual t . The results arc shown in Tabic 5.4. The cr critical energy computed in column 6 is based on ascertaining the mode of instability for a given fault by simulation and then computing the exact u.e.p. evaluating V(x) at this u.e.p. 's
then
obtained by
severe intermachine oscillations due to the postfault network, this u.e.p. coincides with the u.e.p. found from the u.e.p. method as outlined in Sec. 5.4.4. From Table 5.4 it is seen that there is good agreement between The PEBS method has the added it is four times as fast the actual critical clearing time and the critical clearing time using the u.e.p. method as well as the PEBS method. as Lhe u.e.p. meLhod and nbOtt.t repetitive simulation method. advantage of being computationally fasL. In fact,
twenty times as fast as Lhe conventional This enhances the possibility of using for planning as well as 5.10 tor the fault at bus 5.11 shows the swing (i) the VpE()
this method for complementing simulation methods and Vp, F(fi), curves. and Vj^(co) fT(8)'(9 e S ) are given in Fig. Fig.
for online transient security monitoring purposes. Plots of system energies 31 both for stable and unstable cases. It may be observed that,
STABLE tc 0.22
TIME (SECS.)
UNSTABLt
<4
>
LU
LU "ip.. 5.10. Plots of system potential (PK) kinetic (KK) , and total energy s (V) and the quantities F(J) and f_ (0) <G  6 )  classical system representation, cose [31*], 10 unit system, (Reproduced from Ref, [5]),
.189
340 
!3

m)  720
__ I __ I ____1
0.4 0.6
I
1.2 1.4
T1Mb (SEC)
.189
Fig.
5.11.
curveK
case
31*J  10 unit
fiyfir cM n.
For the critically unstable trajectory when VpR(0) v^ reaches a relative minimum and at the
in zero arid
same instant of time, thc mismatch F(8> T (0) ' (C 0""1) changes sign.
Hence,
properties relative to loss of synchronism are confirmed. cr Line Tripped *Faulted Bus 2*3m 4*l4m Critical (sees) Simulation Stable Unstable 0.24 0.2G 0.22 0.20 0.22 0.18 0.18 0.22 0.24 0.22 multiswing figures in b 0.24 0.22 0.24 0.20 0.20 0.24 0.25 0.24 case rackets are Cleari ng tame Controlling UEP 0.235 0.215 0.21 0.23 0.175 0.19 0.22+(0.l5) 0.25+(0.14) 0.23+(0.14> by minimum
*
6 11 15*16 2324* 31 35m
3?m
m +
Critical Energy PEBS Controlling P EBS UEP 0.26 8.99 1 .7 2 0.225 11.75 1 .1 0 0.105 9.31 7 . 95 0. 225 10.47 1 . 1 0 0.185 10.68 7 . 9 0.21 8. 54 1 .5 1 0.21 6.77 6 .35 0. 225 12. 55 1 . 1 0 0.215 10.7 9 .7 energy u.e.p. method
In this chapter, we have discussed the different methods of computing the region of attraction. energy. We mainly considered those methods which took into account the faulted dynamics of the system in computing the critical Methods prior to 1978 which did not consider the fault location linkage established with fact The two numerical examples illustrate the have been briefly reviewed and their conceptual the more recent methods. that Lyapunov function method is stability analysis. Re Terences 1. EiAbiadf A. II. and Nagappan, K.f "Transient Stability Regions tor Multimachine Power Systems", TREE Transactions, Vol. PAS85, 2, Feb. 1966, pp. 169179. Kakimoto, N., Ohsawa, Y and Hayashi, M., "Transient Stability Analysis of Electric Power Systems via Lure Type Lyapunov Function,
accurately the critical clearing time for first swing classical model
2.
COMPUTATION OF STABILITY
REGIONS
Athay, T., Podmore, R. and Virmani, S., "A Practical Method for the Direct Analysis of Transient Stability", IEEE Transactions, Vol. PAS98, 2, March/April 1979, pp. 573584. RibbensPavella, M., Grujic, Lj T., Sabatel, J. and Bouffioux, A., "Direct Methods for Stability Analysis of Large Scale Power Systems", Proceedings IFAC Symposium on 'Computer Applications in Large Scale Power Systems', New Delhi, India, Aug. 1979. Pergamon Press, Oct. 1980. Athay, T., Sherkey, V. R., Podmore, R., Virmani, S. and Puech, C. , "Transient Knergy Stability Analysis", Conference on 'System Engineering for Power: Emergency Operating State Control Section IV, Davos, Switzerland 1979, also U.S. Dept. of Energy Publication N'O. CONF790904PL 1980. Willems, J. L., "Stability Theory of Dynamical Systems", (Book), Thomas Nelson & Sons Ltd., U.K. 1970. Walker, J. A. and McLamroch, N. II., "Finite Regions of Attraction for Problem of Lure", Int. Journal of Control, Vol. 6, Oct. 1967, pp. 331336. Pai, M. A. and Narayana, C. L., "Finite Regions of Attraction for Multinonlinear Systems and its Application to the Power Systems Stability Problem", IEEE Trans, on Automatic Control, Oct. 1976, pp. 716721. RibbensPavella, M., "Online Measurements of Transient Stability Power System Index", in Savulescu S.C. (Editor) "Computerized Operation of Power Systems" (Book), Elsevier Scientific Publishing Co., New York 1976. Uyemura, K. and Matsuki, J., "A Computational Algorithm for Evaluating Unstable Equilibrium States in Power Systems", Electrical Engineering in Japan, Vol.'92, No. 4, 1970,
4.
5.
6. 7.
8.
9.
10.
pp. 4147.
11.
Prabhakara, F. S. and ElAbiad, A. II., "A Simplified Determination of Transient Stability Regions for Lyapunov Methods", IEEE Trans., PAS Vol. PAS94, No. 2, March/April 1975, pp. 672680. Gupta, C. L. and ElAbiad, A. H., "Determination of the Closest Equilibrium State for Lyapunov Methods in Transient Stability Studies", TKRE Trans., Vol. PAS94, Sept./Oct. 1976, pp. 16991712.
12.
CHAPTER V I
V COMPUTATION OF STABILITY REGIONS 193
I
13. RibbensPavella, M. and lieinal, B. , "Fast Determination of Stability Rcqions for Online Power Systems Studies", ProcVol. 12.1, No. 7, July 1976, pp. 689696.
Ii
IEE,
14.
i bbensPave 1 la, M, Howard, J. L. , Lemal, B. and Nquycn, T. I'll., "Transient Stribility Analysis by Scalar l.ynpunov Fund ions: ReccnL Improvements and Practical Results", University ot Liege, Faculty of Applied Science, Report No. 671977. Doraiswami, R. and Konscca, L.G.S., "A Fasl. mid Reliable Dominion ot Transient Stability for Multimachine Powcr Systcms", P.iper A770607 IEEE Winter PES Power Meeting, New York, Feb. 1977. Tavora, C. J. and Smith, O.J.M., "Stability Analysis of Power Systems", IEEE Trans., Vol. PAS91, May/June 1972, pp. 11381145. Varwandkar, S. D. and Prfi , M. A., "Computational Aspects in Practical Application of Lyapunov Method in Power Systems", Proc. IFAC Symposium on 'Computer Applications in Large Scale Power Systems', New Delhi, Aug. 1618, 1979, Perg.imon Press 1980. Tavora, C. > l . and Smith, O.J.M., "Equi 1 ibrium Analysis of Power Systems, IEEE Trans., Vol. PAS91, May/June 1971, pp. 11311137. Cudaru, II., "A General Lyapunov Function tor Multimachine Power Systems with Transfer Conductances", Int. Journal of Control, Vol. 21, No. 2, Feb. 1975, pp. 333343.
15.
1ft.
17.
18. 19.
176
STABILITY
OF
LARGE
BY
DECOMPOSITION
6. 1
I n t roduction
The rapidly increasing size ot power systems and the complex modes of
instabilities that are inherent in such systems has emphasized the need to approach the stability analysis problem in a hicrarchial manner. Such an
approach involves decomposition of the composite system into subsystems which consist ot lower order, isolated (free) subsystems plus the interconnect],ons between the subsystems. mathematically o t for The decomposition Ctin be 110 r
io
rmod
[jhysically.
there arc two approaches possible to ascertain stability of the overall composite system.
consisting of the weiyhLed sum of the tree subsystem Lyapunov functions is constructed and its negative definiteness is examined along trajectories of the composite system. The interconnections influence the negative (WSSLF). In
definiteness of the weighted sum scalar Lyapunov functions the second approach, a vector Lyapunov function
(VLF) whose components The derivative ot the the comparison principle Under
are the subsystem Lyapunov functions is postulated. vector Lyapunov function is examined and applying
certain mathematical assumptions, it can be shown that both the approaches yield equivalent results. There has been a vast literature on the subject of stability of large scale dynamical systems. The concept of vector Lyapunov functions originally systems
CHAPTER V I
J../U
Bailey
[2],
Since then, for nearly a decade there has been extensive 13J and Siljak Ml. In this chapter, our focus
literature on this topic which has been discussed authoritatively in two books, Michel and Miller will be on the application of both the weighted sum scalar Lyapunov function and the vector Lyapunov function approaches to the power system stability problem. we shall Both of these methods share a common principle, namely, the In that context, These range insight such as first study methods of decomposing a power system. decomposition of the composite system into subsystems. from purely mathematical ones to those based on physical coherency, weak tie lines, etc. vector Lyapunov function. We have seen in Chapter V that estimating the region of stability is a very important step in computing the critical clearing time. based methods, systems. 6.2 In decomposition it is attractive to compute the region of stability of the [5] in this connection.
the weighted sum scalar Lyapunov function and the comparison principle based
composite system from regions of stability of the individual free subWc use results of Weisscnbcrger
Decomposition Techniques
Application of thc vector Lyapunov function approach to the transient stability problem of a power system was first initiated by Pai and Narayana [61. In their approach, a nmachine system was decomposed into n(nl)/2 subsystems. Thus, for higher order systems, the most important advantage of the vector Lyapunov function, namely, reducing the dimensionality ot thc stability problem is thus lost. However, as wc shall sec later, their decomposition was essentially an overdescription of the problem. the number of systems. the system has uniform or Later on, a number of other decompositions were proposed which greatly reduced References [7,8,9,10] propose a number of variations of decompositions into (ni) or n subsystems depending on whether
178
VI
STABILITY BY DECOMPOSITION
nonunitorm damping. Reference 110] includes governor action also in the subsystems. These decomposition procedures can take into account transfer conductances in their models. If we ignore transfer conductances, [11,12]. then other types Since the primary treatment to of decomposition such as those preceded by an cquivalencing procedure to group machines into coherent groups arc possible advantage of the stability analysis via decomposition is the possihi 1ity of modelling subsystems more accurately, we restrict our systems which do not simplify the composite model 6.2.1 Decomposition with Nonuniform Damping (3.13) to start with.
Consider a nmachine system whose swing equation is given in Eq. and is reproduced below
M.
d2<s. + D. i i ci P. i  P  ma n ^ j = l n I
J
as. ,r = P.  P , dt i
whore
2 E. C.. 1
11
11
P .
E.  E .   Y. ^ ; cos (6 . .
. .) (6.1)
I D..costi.. x j^l
C. sino. + A 3 JO *J
If the parameters i n Eq. (6.1) correspond to the postfault state, then the postfault SEP flJ can be transferred to the origin and the Eq. be rewritten as Eq.
M
(6.1)
can
(3.35),
i.e.
: ^
+ D
i inr =
 pei<!><
1=1 2
' .......
dt (6.2)
= where A..
^  A.tcosC*.,^) E.IE.IIY..I
 (^j^j)]
^ hihj
"
ij' ^'ij
anCi
^ij
arG re ate<
as
follows:
qj =
EVE.:^ 
E.E.!Y
..sin,.j  A ..Sin*..
(63)
D.. =
iBiHSjlG^ =
e.M
I I
y^ cos*^ = A . .
cos*.. (6.2)
To facilitate the understanding of later material, we write Kq.corresponding to the ntn machine as
VI
STABILITY BY DECOMPOSITION
d26 M n
dL2
41 + D
96 rr = P n dt n3
 P en 
(5) en*1'
V
(i1,2,,nl) and
An.(coS(6nj,nj)  ooS(^r*n.)] D. 1
Define
A .
The
, w. in'
in n
= 1.1
in + M
1 in
n "1'"n A .f .  M. x id ID A. ,f. .
n "Vn
nj nj
" Mn'
A
'Lin "
njfnj
<6'5>
The most eftectivc decomposition which takes into account Lhe transfer conductances in the model is the pairwise decomposition are two possibilities. (i> The choice of subsystem state vector is x^ (g.^gv^r*"^nfn), i1,2,...,ni. We impose the restriction on
> A.
1710].
There
the choice of t.he reference machine that % (i1,2,...,n1). (n1) 18] (ii) [10J subsystems.
i to the reference
o:^ corresponding
In the literature
this is often called the overlapping decoinpos i t i on. Thc overlapping decomposition can be avoided as indicated in Ref. if wc define n subsystems such that thc state vector for the first (n1) for thc nfc^ subsystem
subsystems is x_ = ^in ^in'^n' and the state vector is the scalar u^. We first consider decomposition of thc type In (6.5), the term corresponding to ^n^Anini
M'^A
(i).
an(
. f , = M~ 1 IE.  I K m m n
[oos(6
MY1 . I
1 1
ni*ni' "
cos(6
ni?ni,]
" Mnacin<8in6inin*in
+ M_1D. n (cos6. in cos6"? ) in j n (66)
190
i\,.fJn "
Ilcin(sini'inEin^n
^*
*irr4n
0 0 x.
1
A.
1
0 0
0 X X. ri
A
x. l
(6.8) fi(o)
M lc
i h.(x)
n in 0
a. = 0\x. where.
f
11
i1,2,...,nl
i*i>
sin(o
^ln)
 sin^n
(6.9)
n l 1 1 ~ 1 (M M . I D . *. + . M. A. . f . . )
(M
1
.J"
h.(x) = n
I
' in*i
nj nj
i
M v
A .
M
1 $.
D.
(0.iO ) (6.11 )
200
i n the description
tlie subaystemfi.
(6.8),
This can be avoided by defining the subsystem T S state vectors as x= (fi. 6. ,u>. ) i1,2, .. . ,nl and x = W . r. in in in n n Kow we write the i subsystem description (i1,2,nl) as
VI
STABILITY
BY DECOMPOSITION
A.
i
0
f
i<i>
tJ}.(x))
n1 x n
= >. x
n n j_^ l l
.1  M * t n n
):
A .
(6.12 )
Thc difference between this description and (6.8) is easily seen by inspection. Wc have only second order subsystems to deal with as opposed to third order subsystems. A more general form of this decomposition with asynchronous [9,101. damping and governor representation is contained in Rets. 6.2.2 In Decomposition with Uniform Damping [7]
this case, A...  A for all i. The state vector for the composite system is xT = (6^5^ Vl,n~6nl ,n'i u. . ,. . . ,ic . ). The state space model is given in by
nl,n'
<5,
ic. in in
1 n_1 = w.
iM
>:
in .1  M." f . . i A.
j=1
in n1
A .f.. n (6.13 )
j=1
"3 ^3
13 *3
The subsystem vector is X* = C^^n*S^nr^l and the subsystem state space model becomes
202
1 f Xi ( a. =h . ( x > i) t 1
(6.14) 0 x +
_
0
A i
0V = 11
0)xi
s i n 6 in
l
0
(6.15)
(M
n nj rjj
1 i} 1 M. ) D . d> . + EquaLion (6.14)
~}
1 (M A i
.f .M. irri
n i
1 f,,A..)
ij
(6.16)
ri
where flU
is defined as i n
(6.11).
i s the s i mples t dec omp osi ti on w here the free Two improvements over this decomposition are
0 (6.17)
f.(oi) +
hiM
x. + 0 A.  l 1
X .
1
0.; =
where
f l
0 1 Xj
sin6
f
i<'V
 ^"I'^in'^^i^L*
"
In
lcos (o. +:SS ) j in'  C0S6? ] in (6.18)
in
(MTIM"1)D.
Denoting <Mn1+Mi1)Cin
= w
li
and
W
where
* ^li
+ 1,
2i
^^^f*in>
sin S
tan*. in
(MT1M~1)D. i n
/ (M~1+MT1)C . in n
in
If
^n + (^in
% it/2,
then f"i(cti) if
with the largest inertia, the above condition is likely to be satisfied. (ii) Modification of in i
n
(6.18) in (6.18) " (6.18) and include it in the (6.17) remaining the
In Lhis subsystem description, we delete thc terms corresponding to same, f. (o: j9
W
= "i^in^ ' V O
+ "in^os^+S^)
 cos?nl ) in rin
n
1 Din tan
in
?r =
 d>.
<f^n being the angle of the transfer conductance between internal buses i and n. Under these conditions h^ (x) becomes h.(x) =
l^n^^'^i^ln^J^^^A^
J:
1 (M X A nj n j
1 .f i .M.
13
A..f..)J
13'
(6.21)
Wc thus see that there arc quite a few variations in constructing thc free subsystem structure in pairwise decomposition techniques.
^04
POWKK
KYSTKM
H TATSLLL
11
6.2.3
VI
STABILITY BY DECOMPOSITION
(i) In a pairwise decomposition structure, we can in addition take into account asynchronous damping and first order proportional governors [10].
(ii) One of the restrictions in the nonuniform damping case is that A > A.. This is necessary in order that the linear part of the free subsystem in (6.8) be stable. This restriction can be removed if instead of u*n we introduce wQ which is the velocity corresponding to the center of angle. A new set of decomposition and its variations is possible [14]. (iii) If we do an apriori coherency analysis according to some criteria and group the machines, then the number of subsystems could be reduced. Lyapunov function can be applied to the reduced model. discussed in Kefs. [11J and the scope of our discussion . (iv) All the px~evious techniques rely on a mathematical decomposition of Since the description is based on Lhe bus admittance However, if a This technique into 112] however does not strictly fall
matrix at the internal nodes after eliminating all physical nodes, any type of physical decomposition is therefore ruled out. structure preserving model is used as in Ref. decomposition is possible 116]. Although a great number of decomposition techniques have been discussed in the literature, there is no clear cut evidence to indicate that any one method is preferable. Since our main focus is on methods which have met with some success in practical application, detailed discussion of decomposition techniques has been limited to such cases only. oi i S Weighted Sum Scalar Lyapunov Function Approach H5], then a physical
We first state some general results before applying it for the power sysLem problem. and MillerThe treatment in Lhis section follows largely that [3] . in Michel
x =
k subsystems with x^ of dimension n^ as the state vector of the i*"* 1 subsystem. Hence,
I
1=1
n.  N
1
T
.
and
x  (x 1 ,x 2 , ,x k )
Fret the decomposition have the form x. where g^(x) = f^lxy + g.(x) i=l,2,...,k (6.23) A
m
ki
q.tx)
= f i (x i )
(6.24)
represent.s the i*"* 1 free subsystem. The following theorem due to Michel [17] is fundamental to the weighted
sum scalar Lyapunov function approach. Theorem 6.1 The equilibrium x = 0 of the composite system (6.23) following conditions arc satisfied. (i) Each free subsystem (6.24)
v
is a.s.i.l. if the
^' x ^^
11 2
1
(6.25) i
where
 * 
(ii) Given v^(x^), there exists constants a^. such that the followinq inequality is satisfied with respect to the interconnection terms
k VV<x ) g (x) < x  I (6.26) a . x 1 1 1 j=l 11 (m) Given o . of hypothesis (i) there exists a 1 kvector T a = (aa ) > 0 such that the test matrix S = ls..J ~ 1 K 13 specified by
a (
i i+aii'
206
s1 =
^ (6.27
)
(a a
i ij+ajaj.i.)/2 is
negative definite. Proof: For the composite system, choose the Lyapunov function k V(x) Clearly V(x) fact Lhat have V(x) =
k
% i=l
1 1
ct.v. (x.) ~1
> 0 in Lhe entire region by virtue of hypoLhesis (i) > 0 (i=l,...,k). Along the solutions of
(6.23)
+ g.(x)1)
k
&
Q LA  0 , 1 i1 =
X, I
x.  I 1 j = l
a..x.l
13 3
(6.29)
and R ij
[r..]
be the k*k
specified by
11
ij
*
a
i^.j
m
T
2
We now have T T 8+ w R w = w (
V(x)
<
) w = w S w <.
1,
l l 2
where S is the tost matrix of hypothesis (iii) and A^ denotes the maxinium eigenvalue. Hence V(x) is negative definite if S is negative definite. also note that since S is symmetric, all its eigenval ties are real. Since S is negative definite, *<> < 0. We
ll
(1 >
y!il
0.4 Region of Attraction Suppose that the hypothesis state space.
VI
STABILITY BY DECOMPOSITION
207
asymptotically stable in a region around the state space and the region of attraction is defined as v.(x.) vQ^. Then, as shown by Weissenberger of attraction of thc composite system is given by V(x) Y Min whore (u.v .) y i or
15],
the region
(6.33 )
Ki<k
This rcqion of attraction is in general conservative and one way ot improving it is to optimize the "^'s, a topic to be discussed later. 6. Results Involving MMatrices
It may be observed that in the application of Theorem 6.1, wc need the existence of suitable cu's > 0 (i=l,2,...,k). I n the special case when (nonnegative) we to establish [3] thc off diagonal elements of the test matrix S are > 0 can utilize properties ot the socalled Mmatrices results which are easier to apply because of weight age vector ii and (ii) (i)
equivalent to those obtained by the comparison principle and vector Lyapunov function. Wc first review certain properties of Mmatrices. MMatrices [3] Dof in ition A teal kk matrix D (i) and d. . (ii) < 0 i^j ] is sa.id t.o be an Mmatrix if (i.e. all off diagonal elements of D are nonpositive)
determinant dn ....
a
=1,2,...,k (6.34 )
h:l
Mmatrices. a) D is an Mmatrix.
tl
0 such that Du
0. T
d) The real parts of all eigenvalues of D are positive. e) D is nonsinqular and all elements of D all diagonal elements of p From (b) (i) and (c) are nonnegative (in fact, are positive). respectively that (jl,...,k) such that
it also follows,
.a. >
(ii)
VI
STABILITY BY DECOMPOSITION
207
Another useful property of Mmatrices which we require is the following. Corollary. elements, matrix B  DTA + A D is positive definite. This concludes the discussion on Mmatrices and its propcrtic! Wc now reformulate the stability Theorem 6.1 Michel and Miller Theorem 6.2 [3] (i) and (ii) of Theorem 6.1 are true with the \ 3]. in terms of Mmatriccs following (6.35 ) Let D i.e. d^.. Id^J be a k*k matrix with nonpositive off diagonal < 0 i#j. Then D is an Mmatrix if and only if there
(i) The successive principal minors of the k*k Lest matrix D = all positive, where i " aii d. = .
1=3
(6.36 )
lj
(ii) The real parts of Lhe eigenvalues of D are all positive, (iii) exist positive constants k (o,a..)
1 11 1J
There
(i=l,2,...,k)
>
7.
i=l,2,...,k
Proof. Hence,
Since by assumption a^.. > 0 tor i#j and since the successive from the property of Mmatriccs, there exists a diagonal matrix A
s
principal minors of D are positive, it follows that D is an Mmatrix.  diag (,,..., a.) that. 2S ) is positive definite. DTA + A D (6.39 0 such " 1
K
210
Hence 2S 
2 Is. where
ft ( a +a
i  i ii>
i=3
(6.40 )
i
is positive definite. If 2S is positive definite, is negative definite. of Theorem 6.1. (iv) Hence, (ii) are true since D This tantarnounts to satisfying condition (iii) the equilibrium is a.s.i.l. Conditions is an Mmatrix. liquations (6.37) and (0.38) are referred to as row dominance and column dominance conditions, respectively. It may be noted that condition (i) is independent of c^'s. Also, and 6.6 (6.38) (6.37) can be used to optimize the region of attraction as shown below. Optimizing the Region of Attraction
The region of attraction for the composite system has been defined to be V (x) where < y in (a.v.) , 1 oi
.
To get the best estimate, we need to optimize a^'s. a.'s with cither
A .'s
or n. s in
(6.37)
and
(6.38). k)
it follows from
(6.37)
that
following two criteria for optimizing the region of attraction. Criterion 1 Minimize the trace of the test matrix S Min z = a k
subject to
(6.37)
=1 k namely
n
(6.42)
I a. = 1 i=l Criterion 2
Maximize the weighted sum of the boundaries of the stability region of the subsystems
Max
ZV
a v
i oi
(6.43)
(6.191)
210
o subject to
iI
(6.42).
The
Let tx^ be the optimized value of ft^'s by either criterion 1 or 2. region of stability is defined to be
I
7.
*
a.v. (x.)
1 1 1
< V
)
il and V6 =
(6.44
In view of the numerous equivalent properties of Mmatrices, there arc quite a few other possibilities for optimizing the n .'s. l
6.7
[13]
Since the best results to date have been obtained by the weighted sun scalar Lyapunov function method using decomposition, we present a numerical example in detail. We consider a uniformly damped system and first state the general formulation for an nmachine system before presenting the example. The presentation follows closely that of Chen and Schiny.iriger [13], The subsystem modelling is the same as in
ii
&
W*
i(x)
ip
(6.45)
(6.192)
212
in
x
ii
II
where 
in  6?
0 A. 1 A
2i
U
u> in
T
b 10 fi(ci)
th machine having the largest inertia is designated as the n machine and 1 Hence M" << 1 n KT chosen as the reference machine. 1 so that in (0.19) is almost equal to  t). is where 6. T J in ' ~" ~ * 2 in the transfer admittance angle between machine i and n. 6.7.1
1 1
and hi<x)
are defined in
(6.18)
and
(6.19).
The
Subsystem Analysis
*  ii
fJ
6 fiK
i1,2,...n1 (6.40 )
i = ^i*i
i1^
in Sec. 4.6 since Eq. (6.46) is analogous to Eq. (4.37) with the followinq
= M 1 D /M =
A
M J,J li
in 2i
eg in (6.47 ) (4.44).
+ P
i i>
li
2 4
? x
' lix2i
2 .
+
v
2i>
'o
is evaluated as
i'i^
;1
lenq
(6.46)
(6.193)
vi
2 =  2
STABILITY BY DECOMPOSITION
213
. A fi(xli)
xu
2
X
X li ' 2 2i Lhe
X
(6.49 )
is in a region around f (x i ii> origin, we get an upper bound on   as 3 where x li >fl(*li> ax ii
v n x^^=0
Hence
i*xiJ
 ~ii'
where
(6.50 )
where
ci
The region ot stability of the free subsystem is given using Eq. and the equivalences (6.47) as
in
(6.51)
f.(o.)da.
V(x) =
a.v.(x.)
11 1
1*1
n1 X {a ii
V(x)
1 1  1
li *
lk1 ,X li
X
+ x
and
lk *X2i'
and
li
 Hi
[71
(6.194)
195
co8
We majorize (6.52).
h. (x) =
h.2(x)
(6.55 )
nj
nj
.)
 oos(6nj$nj)]
n 1
I .1 M. (6 . .d>. .) i ID
Now
]_j_",'x2i^'1i2
(1 4 fjlM"1
V
i n]1 A
Aijsin(6^j*ij)xlixlj;
(6.57)
+ M
1 n
n x
"
} x. '"l"
2 _. sin <<*?.$. .)
Z  n1
1
n1 (1 + ^{M,1
A. .1
13  i j
"1
! X .
j=l n1
ij~Tij"
*i*j
^ M"
' Ir ix .
'_j
A . I
n j
$ in
1
n j
(6S
1
nj
" l
n1 n1
a.o.x.
Z
1=1 ]=1
(6.59)
1=1
(6.58)
*3
VI
STABILITY BY DECOMPOSITION
mm
au =
M^
Aijsin(.,pij)
ii 
(1 +
1'
{m 1 R
8in(S nji
r*i
(6 601
* M _1 A. . I sin
. .) I)
i^j
a'^g
Also, (iii)
of Theorem 6.2 to ensure asymptotic stability of the composite system, the (n1) x (n1) test matrix being f> = [tl^j] where i=i J (6.61)
r o.  a..
d. . =
\
ii
13
In particular, we utilize the row dominance property, > 0 (i=l,2,...,nl) such that n1 7. a.a.. j _j j
w. (a.a. . ) l I 11
ij
> 0
i=l,
2,.
. . ,nl
(6.62)
This implies c.fc^a^j > 0 and since > 0, we have > a^, i.e.
1
>
(1 A 2>
M
+/ V
13
13
(6.63)
Equation
(6.63)
as the prefault state of the individual subsystems should be such as to overcome the factors arising through the network interconnections Y^_.. equation also reveals that the weaker the value of Y is for the c.  a.to be > 0 which is only a sufficient l This ii
condition.
such a mariner that the resultant subsystems are weakly coupled the decomposition based on the internal node description, v^j's hence there is very little flexibility. on structure preserving models 6.7.3 Numeri cal Example [15] [1.3] might be helpful.
In
fixed and
6.16.6 give the pertinent data. Table 6.7 shows that four different fault locations are examined. For example, in Test No. 1 we assume that a 3phase The numerical results are adapted from fault occurs near bus 8 and that line 89 is tripped after fault is cleared. We assume A = .001 and fJ = 1.0. Chen and Schinzinger [13].
0. 2
O A
PiK' 6 . 1
S i n g l e l i n e d i a g r a m o f f o u r m a c h i n e power s y s t e m .
VI
STABILITY BY DECOMPOSITION
From Bus
To Bus
P.U.
Impedance
5 5 5 6 7 7 8
Table
6 7 10 8 10 9 6.1
Generator Number
MVA Capaci ty
1
2
3 4
NOTES:
15 40 30 100
1. M A  4 nfH i 2.
D^ = 2 rfd i
P.U.  rad/rad
3.
4.
h\
6.2
Machine Constants
1 0 2 0 3 0 4 0
0 0 0 0
0 0 1 0
0 0 0 2
Table 6.3 Reduced Bus Admittance Matrix for Postfault System (line 89 tripped)
VI
STABILITY OF DECOMPOSITION
2 3 4
0 2.0A9 .CT 0 0
Tabic 6.4
Generator Number
E.
6 . x in
1.057/5.30" 1.155/11.08 1.095/5.48 1.000/0.08 Table 6.5 Internal Voltage of PreFault System E. 6. l in
Generator Number 1.057/5.69 1.152/14.39 1.095/2.27 1.000/0.08" Table 6.6 Test Number Fault Location Line Tripped Criterion I L'ritical Criterion 2 Switching rime in sec RnngeKntta Method
Internal Voltage for PostFault System 1 2 3 4 Line 89 at Line 89 at Line 57 at Line 57 at Bus 8 Side Bus 9 Side Bus 7 Side Bus 5 Side Line 89 Line 89 Line 57 'Line 57 0.36 0.38 0.25 0.30 0.36 0. 38 0.30 0.37 0.36 0.40 0.35 0.44
Table 6.7
VI
STABILITY OF DECOMPOSITION
199
With machine 4 as reference and letting = 1, i1,2,3, the following p matrix is obtained using 3.
D =
16.60)
with
.0005.
10
(6.64 )
It may be verified that D is indeed an Mmatrix. equilibrium is asymptotically stable. we compute vQ^ subsystems. s li dx dF
2i
x
Hence
the
(i=l,2,3)
the stability regions of the three free the free subsystem for machine 1
To illustrate,
2i
dx dt
li
0j_
[1
x
0]
2i
The nonlinearity f^ (a^) satisfies the condition c^f^(c^) > 0 in the region it _2^in+^in) < i * 71 " 2 ( < S tn +*inJ ' i"e' 3.298 o, < 2.984. Using (6.47) the stability region vm is
<
3
02
01 Similarly the region for machines 2 and 5.477*10"4 and vQ3 = 9.281*10~4
3 are computed as v
Optimizing a^'s and Computing the Region of Attraction 's>. Using Criterion 1
We use the results of section 6.6 to optimize the rcqion of attraction by an Appropriate choice of weight ages gives Min z (3.0130^4.01o2+3.976o3)lG
4
VI
STABILITY OF DECOMPOSITION
200
(6.67) Subject to
a
+ a
+ a
3 ~ *
3.013^  0.733v*2  0.580a3 > .0 (6.68) 1.431a] + 4.01a2  0.623a3 > 0 0.757^  0.456a? + 3.9763 > 0
Min
= 0.6139, a.. 0.2415, a, = 0.1446 and 5 * * (a. v .) turns out to be 1.32*10 . Hence the
oi
_4
1
For this particular postfault condition, the solution obtained by using Criterion 2 yields the same as tt^'s discussed in Chapter V is 0.36 sec. actual simulation is by both the methods. Table 6.7 summarizes the list of results at the several fault locations in this four machine system. It appears that Criterion 2 gives the results The discrepancy is This which are less conservative. However, by both criteria the results agree very well by actual simulation method for tests 1 and 2. larger for tests 3 and 4. This means that the decomposition technique gives in Criterion 1. by The critical switching time determined using the procedure The value of found to be 0.37 sec. Thus there is a good agreement
favorable results only for certain faults and not for other ones.
is not a desirable feature and more research is needed to remove this drawback. 6.8 Vector Lyapunov Function Approach Using the Comparison Pr.inci pie
Historically the vector Lyapunov function method using comparison principle was the first technique to be used for stability analysis of large scale systems via decomposition [l,2j. In one sense, it is a very general approach and has naturally generated a lot of interest in control theory and mathematical circles. considered thus far, However, for most of thc specific cases
function method where usage of the comparison principle is not required. In power systems, comparison principle based vector Lyapunov function (VLF) was used initially together with the requirement that the free [0,7,8]. This condition is, subsystem be exponentially stable results were extremely conservative.
however, stringent so far as power systems ate considered and hence thc Therefore, the method did not In this section, we result in practical applications to realistic systems, in spite of the relaxation on the conditions on the subsystems [9]. problem. ture [3] . briefly review the VLF method and indicate the application to the power system The comparison principle is discussed extensively in the literaOur presentation is along the lines of Michel and Miller [3,4].
4
201
6.8.1
i =
Definition
(6 69 )
H(y) is said to be quasimonotone if for each component H^ 1=1,2,..,J( the inequality tij(y) < tu(z) is true whenever y^ < the condition for all ij'j and y^ > 0 (ij^j). If H(y) 
< y(t ),
< y(t)
for
t > tQ
(6.71)
to hold is that H be quasimonotone. In the case H(V)  P V, the condition reduces to p.. > 0 (i/j). then y(t) and r(t) > 0 as t > #i For the
linear case it also follows that if P has eigenvalues whose real part < 0, This, together with the property that p^ > 0 (i/j) is equivalent to P being an Mmatrix. We now state the theorem for a.s.i.l. of the system (6.23) using the comparison principle and VLF. Tjicqrem 6 3 Let the free subsystems v.(x.) (6.24) < 3 $ i have the Lyapunov function v^(x^) and let v i i as (X
), (6.24
(v^
(x1),...,v^(xk)) v. (at. )
1 _:L
. =
Now [Vv. (x.) ]T[f  (x)+g. (x. )1 1 _1 * "x " x 1 + Vv.(xi)Tgi(xi) lp^.], p.. (6.72) > 0 i^i such
(6.23)
 fii vi(xi)
T.
13
1 1
p, . v, (x. )
VI
STABILITY OF
DECOMPOSITION
202
(6.23)
p...
Hence
If the matrix P has eigenvalues with negative real parts, the equilibrium of the composite system is asymptotically stable. Proof: The proof involves using Theorem 6.1 concerning The two assumptions, From the
p.. ^
condition
11 (1)
P
1 1 y%l
(ii)
and
equate
.'S with
,...,a )
(6.75 )
V(x) 
J! ii=i
a.v. (x.)
V(x)
< a P V(x)
Since < ^ P
<
asymptotically stable.
linearized version of the comparison principle, the VLF and weighted sum scalar Lyapunov function are really equivalent.
9
It has been pointed out that the VLF method was the first one to be applied to power systems using this method. inequality V(x)
[6]
and hence there is a good amount of literature it is generally necessary to have vjtx^) which Hence, it poses a problem in power system
< P V(x),
v
application where ^tx^) consists of a quadratic plus an integral of thc Ways of circumventing this difficulty often resulted in We briefly sketch a number of approaches to very conservative results.
VI
STABILITY OF DECOMPOSITION
203
6.9.1
Approach Based
011
Exponential Stability of Subsystems (20.7] With each subsystem we associate where v^(??^) is given by (6.48).
^{x^ ) k
v^^(x^)
Each subsystem is assumed to be exponentially stable with the following estimates on v^fx.) and v^(x^). 01
II
^1
II
1
< rii2
(6.77)
{W
<x .)]Th
(x) 
<
j=l
=
{v V(x)
p.. = 13
^iinIl
(6.79)
In
(6.77)
and
(6.78)
Hence P in
(6.79)
Let thc region of attraction of the free subsystem be defined by v. (x.) Define v? = y$. > is defined by v(x) where < y Max jv. (x.)/p. 1 < i< k (vV /p.)
1 1
(6.81)
v(x) =
and
f 
Min Ki<k
p is computed using the properties ot P which is an Mmatrix (see Sec. as follows: elements of P positive). by P Since P is an Mmatrix, P~ are nonnegative
6.5)
The application of this method for a power system is briefly sketched below for the uniform damping case given by function from (6.48)
v
(X;L'
VI
*i Si
where
Vxi>
2i ^ Si
(6.83 )
Si
1
\
and
T
2 where as
v (x
i i> K "xi
2i
, the numbers
(6.84 )
Since v. (X^J as
vV (x.)
2
r)^,
"il
ri3 where A /2
"If2^'
m l
*12
>M/2<i>' (6.05 )
^ (;
and X_, are the minimum and maximum eigenvalues of the M The C ^j's are computed using a procedure similar to
s
indicated matrices.
[71.
This methodology has been extended to Lhe nonuniform damping case in Ref. 18] with the stability region defined as n1 7. i=i a. v. (x.)
1 1
< v
cc^'s are chosen to satisfy {6.39) with D P i.e. where T P A. + A P is positive definite with A = Diag (a.). y is computed as Min l<i<nl (<*.v).
1 1
reduces to that~of the weighted sum scalar Lyapunov function method. However, the majorizing process is different in the two oases. The VLF This majorization seems to be more restrictive than the WSSLF method. perhaps explains why the results are better in the latter case. 6.9.3 Combined VLF and Weighted Sum Scalar Lyapunov Function Approach 19]~
i n the weighted sum scalar Lyapunov function method, we chose majorization on v.{x.) to be v. (x. ) < o. x.  'x.' . Instead of , we can choose other functions such as positive
VI
so that v^(x^)
The
analysis can proceed along the same lines as in Sec. the weighted sum scalar Lyapunov function method we imposed no sign restrictions on a^jsystems, Instead of considering c. and a.. l
what is called a "oneshot" aggregation by assuming the existence p.. that vi(x.) * tgrad v. (Xj^) ]Tlfi(xi) k < I p. . vi. (x. )u. (x.) +hi<xi)]
(6.86
We can now restate the stability theorem, 6.1, as follows [9]. Theorem 6.4 Given v.(x.)
X 1
and u.{x.)
J.
satisfying
rp
(6.86),
if
(a^,...,d^)
P B + B P is negative definite, then the origin of (6.23) is asymptotically stable. i n the above theorem, p^ may be of any sign. A closer look at this theorem reveals that it is equivalent to Theorem 6.1 with
V^i*
H%ll
+ a i=j
Furthermore, equivalent.
p B + B P = 2S.
1f in addition P is an Mttiatrix,
of
a.
The application of thc preceding theorem to power systems has been done in Ref. (i)
[10J
in addition to mechanical damping has been taken into account. (ii) Governor action is incorporated for each machine. ( i i i ) The decomposition is done along the lines discussed in 6,2.1(ii), there is no overlapping decomposition. (iv) Thc region of attraction is obtained as the union of Lhe two reqions, i.e.
one obtained by weighted sum scalar Lyapunov function method and the other as in VLF method. However, the computation results reported are quite conservative. reason is the type of ma jori ?.at ion used which removes the fine physical choice of j_' s
The main
information about
VI
Conclusions In this chapter, we have presented a systematic methodoloqy of stability analysis by large scale power systems by decomposition. As shown, the
weighted sum scalar Lyapunov function approach i s equivalent, to thc vector Lyapunov function
Further research in large scale power system stability should be directed towards improved methods of decomposition retaining the physical integrity of the system and optimizing the region of attraction.
VI
STABILITY OF DECOMPOSITION
207
References 1. 2. Bellman, R. , "Vector Lyapunov Functions", SIAM Journal of Control, Vol. 1, 1962, pp. 3234. Bailey, F. N., "The Application of Lyapunov's Second Method to Interconnected Systems", SIAM Journal of Control, Vol. 3, 1966, pp. 443462. Michel, A. N. and Miller, R. K., "Qualitative Analysis of Large Scale Dynamical Systems", (Book) Academic Press, New York, 1977. Siljak, D. D., "Large Scale Dynamic Systems  Stability and Structure", North Holland, New York, 1978. weissenberger, 5., "Stability Regions of Large Scale Systems", Automatics, Vol. 9, 3979, pp. 653663. Pai, M. A. and Narayana, C. L., "Stability of Large Scale Power Systems", Proceedings of 6th IFAC world Contress, Boston, 1975, Paper 31.6, pp. 110. Jocic, Lj. B., RibbensPavella, M, and Siljak, D, D., "Multimachine Power Systems; Stability, Decomposition and Aggregation", IEEE Trans., vol. AC23, 1978, pp. 325332. Jocic, Lj. B. and Siljak, D. D., "Decomposition and Stability of Multimachine Power Systems", Proceedings Of 7th IFAC Congress, Helsinki, 1978, pp. 2126. Gruijc, Lj. Ti and RibbensPavella, M., "Relaxed Large Scale Systems Stability Analysis Applied to Power Systems", Proc. 7th IFAC World Congress, Helsinki, 1978. RibbensPavella, M., Gruijc, Lj. T. and sabatel, J. , "Direct Methods for Stability Analysis of Large Scale Power Systems", Proc. of IFAC Symposium on 'Computer Applications on Large Scale Power Systems', New Delhi, India, Aug. 1618, 1979. Gruijc, Lj. T. , Darwish, M. and Fantin, J., "Coherence, Vector Lyapunov Functions and Large Scale Power Systems", International Journal of Control, Vol. 10, 1979, pp. 351362. Araki, M., MohscnMctwally, M. and Siljak, D. D., "Generalized Dccomposions for Transient Stability Analysis of MulLimachinc Power Systems", Proceedings of 7ACC, San Francisco, Paper No. TA3B, 1900. Chen, Y. K. and Schinzingcr, R., "Lyapunov Stability of Multimachine Power Systems Using DecornpositionAgyregation Method", Paper A800364, IEEE Winter PES Meeting, New York, Feb. 1980. Vittal, V., "Lyapunov Stability Analysis of Large Scale Power Systems", M. Tech. Thesis, I.I.T., Kanpur, July 1979.
3. 4. 5. 6.
I.
0.
9.
10.
II.
12.
13.
14.
lb.
Bergen, A. R. and Hill, I). J., "A Structure Prcscrvinq Model for Power Svstem Stability Analysis", IEEE TranG., Vol, PAS100, Jan. 1981, pp. 2535. Pai, M. A., "Stability of Large Scale Interconnected Power Systems", Proc. IEEE International Conference on Circuits and Computers, Oct. 13, 1980, Port Chester, New York, pp. 413416. Michel, A. N., "Stability Analysis of Interconnected Systems", SIAM J. Control, Vol. 12, No. 3, Aug. 1974, pp. 554579. Michel, A. N. and Porter, D. W., "Stability of Composite Systems", Fourth Asilomar Conference on Circuits and Systems, Monterey, California, 1970. ElAbiad, A. ll. and Nagappan, K. , "Transient Stability Region for Multimachine Power Systems", IEEE Trans., Vol. PAS85, No. 2, Feb. 1966, pp. 169178.
16.
17. 18.
19.
20.
CHAPTER VI3
CONCLUSIONS
7.1
Introduotion power
system stability analysis may remain largely as a theoretical development, the picture has brightened considerably as a result of recent research efforts in the computation of stability regions. This was discussed at length in Thus Chapter v where fault location was taken into account in computinq t cr.
the direct method of stability analysis of pov/er system will continue to be an active area ot research in the future with greater emphasis being put on application and online implementation for transient security evaluation. Three broad thrust areas of future research can be identified and we will briefly review the v/ork in each one of them. 7.2 Structure Preserving Models
The stability analysis so far has been done on the baiss of converting the loads into constant impedances and obtaining an internal bus description after eliminating all the physical buses. besides modelling the loads inaccurately, this approach masks the topology of the network, preventing a better understanding of energy and power transfer during the disturbance period. An effort to move away from this viewpoint was first proposed by [1]. Their model includes frequency dependence of the Bergen and Hill
loads and retains the topological structure of the transmission network. There is a first order differential equation for the phase angle tor each physical bus and functionally the equations are of the form
x = F(x,y> cy = C(x,y)
(7.1} (7.2)
where x is the state vector associated with the rotor angles and speed deviations and y is the state vector associated with phase angles of the physical buses. r. > 0 is a small parameter identified with the damping (7.1) and (7.2) can be cast in the [1,2]. coefficient at the loads. Kquations
Lure form and the multivariable eopov Criterion can then be applied
209
2 32
terms of choosing physically realistic parameters for the load model. Using the philosophy of structure preserving models, voltage dependent nonlinear static loads have been Incorporated and demonstrated successfully on a practical system by Athay and Sun Equations (7.1) and (7.2) [3].
are in the singularly perturbed form and with Sastry and Varaiya
c  0, we obtain a degenerate system or an implicit set of differential equations whose stability is difficult to investigate. f4] degenerate and the singularly perturbed system. have examined the local stability of the equilibrium points of both the They have also used the
formulation for the hierarchial stability analysis and alert state steering control of interconnected power systemsPractical interpretation and validation of some of these ideas yet remain to be done. It is therefore clear that structure preserving models open up altogether a new area of research for power engineers. Since transmission lines are terms in the Lyapunov [i>] , the modelled as such, the line resistance can be neglected in comparison to the reactance thus avoiding the path dependent integral to incorporate livuc links function. Another advantage of retaining the network structure is the ability in the stability analysis. In Ref. concept of distribution factors is used to reflect both the loads and power in the DC link at the internal nodes ot the generators as additional power injections. The transmission network is then reduced to the internal buses A modified energy
function approach is then applied to the system containing the machines and the DC link dynamics. Results of fc^o are quite encouraging using this method and extension to systems containing multiterminal HVDC links is a possibility. 7.3 Stability by Decomposition
In Chapter VI, the analysis was based on thc pairwisc decompcition of a large scale power system. This does not take advantage of apriori information (6J have proposed a generalized decomposition However, it The about the system such as weak coupling, strongly connected machines, etc. Araki, Metwally and Siljak where more than two machines may be included in a subsystem.
is still necessary to include the reference machine in each subsystem. VI. Alternatively if we use a structure preserving model, of physical decomposition exists tearing techniques (Diakoptios) is yet to be demonstrated. 7.4 Security Assessment (Ref. [16], Chapter
analysis is done on the basis of the internal node description as in Chapter the possibility IV) using Kron's validation
2 32
In spite ot narrowing the gap between the critical clearing time obl.ained by simulation and Lyapunov's method (using classical model), it is unlikely simulation methods for that Lyapunov's direct method will replace existing
planning purposes. The reason is simply that for most of the problems in
planning
studies, one requires detailed models of the generating unit (i.e. Nonce at best
synchronous machine, excitation system and the governor). Lyapunov functions for such complex models are not available at the moment. Lyapunov's method will compliment the simulation method in planning studies
by filtering out from the large number of cases being studied, those which require detailed scrutiny. However, role. it is in Lhe area of LransienL securiLy moniLoring where References [3740] of Chapter suggest the possible use of However, this subject
computational times are very demanding that Lyapunov's method can play a major Lyapunov function in be developed.
CONCI.US I ONS
212
References 1. Bergen, A. Ft. and Hill, n. J., "A Structure Preserving Model Tor Power System Stability Analysis", IEEE Trans. PAS100, Jan. 1981, pp. 2535. Pai, M. A., "Some Mathematical Aspects of Power System Stability by Lyapunov's Method" in: Erisman, A. M., Neves, K. W., Dwarakanath, M. H., (Editors) "Electric Power Problems, The Mathematical Challenge", (Book) Society for Industrial and Applied Mathematics (SIAM) Philadelphia, 1980. Athay, T. M. and Sun, D. I., "An Improved Energy Function for Transient Stability Analysis", IEEE International Symposium on Circuits and Systems, Chicaqo, April 1981. Sastry, S. and Varaiya, P., "Hicrarchial Stability and Alert State Steering Control O f Power Systems", IKKK Trans. Circuits and Systems, Vol. CAS27, 1980. Pai, M. , Padiyar, K. R. and Radhakrishna, C, "Transient Stability Analysis of Multimachine AC/DC Power Systems via Energy Function Method", Paper 81 SM 4084, IEEE PES Summer Power Meeting, Portland, Oregon, July 1981. Araki, M., Metwally, M. M. and Siljak, D. D., "General i/.ed Decompositions for Transient Stability Analysis of Multimachine Power Systems", Proceedings of Joint Automatic Control Conference, San Francisoo, Aug. 1980. Uappj 11. Systems", H., "Picccwise Methods and Application to Power (Hook) John Wiley, New York, 1980.
2.
3.
4.
5.
6.
7. 8.
Grujic, I.j. T., narwish, M. and Fantin, J., "Coherence, Vector Lyapunov Functions and Large Scale Power Systems", International Journal of Control, Vol. 10, 1979, pp. 351362. Mahalanabis, A. K. and Singh, K ., "On the Analysis anc Improvement of Transient Stability of Multimachine Power Systems", IEEE Trans. PAS100, April 1981, pp. 15741580
9.
APPENDIX I
A1
Introduction
A knowlcdqc ot the synchronous machine model is essential for understanding the various assumptions implicit in thc simplfied classical model generally used in transient stability Studies. Future research in Lyapunov stability analysis ot power sysLcms will undoubtedly try to relax some of these assumptions and accomodate a better model of the synchronous machine. The brief review which follows derives three types o f models. (i) (ii) The two axis model with a damper winding in each axis. The one axis E^ model with no damper windings but including the flux decay effects. (iii) A.2 The classical model. Model in Park's Variables
The synchronous machine consists of three phase windings on the sLalor and three windings on the rotor, namely a main field winding along thc direct axis and two fictious short circuited damper windings, one each along the direct (d) and quadrature (q) axis respectively (Fig. A.l). We adopt thc convention of the d axis leading thc q axis following Ref. A coupled circuit viewpoint of the six windings differential e q u a t i o r i H indicates Lhat thc
11].
to adopt the well known Park's transformation which converts the three stator windings into two equivalent fictious windings called the daxis and qaxis windings, moving synchronously with the rotor. the After thc transformation,
2 213
238
POWER
SYSTEM STABILITY
a AXIS
A.l. machine.
QUADRATURE AXIS
MulLiporL representation of Xi characteristics in direct and quadraLure axis. (2 stator and 3 rotor)
is a
will now have only timeinvariant coefficients. Pack's transformation power invariant transformation and Kef. derivation.
11J
We denote the trrttisformed st^Lor windings as d and q axis the direct a x i s damper winding
238
POWER
SYSTEM STABILITY
as I) and the quadrature axis damper winding as 0. winding model is shown i n Fig. A. 2.
the
Kirchoff's
voltage law equations for the five windings and s i n c e the stator windings also rotate synchronously with the rotor, there is no mutual coupling between thc direct axis and quadratic axis windings. induced voltage eguations dX .
V ri
the
(A.l)
V q = ri .
 as* + w X .
d
dX
dt d
Rotor Equations
ax
v
f *
Vf
ar
d X
Tir
(A. 2)
d>.
1 ) 0
= nD
r x
0=
Ti
l' s represent the resistance and X's the flux linkages of the respective windings.
be wr
L
AD
\i
1
L L
AD AD
L
' AD L
AD A
1
(A.3)
240
q h. (A.4) Q q
AO L I..
+
Al F AD) F' D AD This assumes a common flux path and s for each I of the two axes. Equations (A.3} and (A.4) 0 can be viewed as a multiport representation as i n Fiy. A. 3. Together Eqs. (A.1) (A.4) can be given the equivalent circuit repreThe sum of the outputs of the two
From
sentation as in Fig.
A.4.
dependent voltage sources represents the electrical output P^ of the machine, i.e., P^ = ^^d^q~^q^"d^w' and (A.4), [2]
T
MDO
*a F MDO
MDOSF
C1
T
MDO D F MDO F
MDO "D
r
\i i
_ F
MQO+4q
MQO Q
(A.5)
(A.6) "Q
sl
u
,1 Q
HQO Q
&2 MQO
'0
1 where
MDO
L
AD *D
*F
and 1
MQO
+ 4
k
i
It is customary to define the socalled subtransient inductances in the direct and quadrature axis as L^' = + ^ and I/' L,,^_ + f. respectively. 1 In the absence of damper J q MQO q
240
windings, we have in the direct axis, the transient inductance i.l Ail these inductances are interpreted in Fiy. Av5 as an aid for quick reference. With these preliminaries, we now proceed to develop the model using the hybrid characterization (A.5) and (A.6).
(1/ L , _ + l/f.)1 + 9 and the steady state inductance in Cl A1' r AO + *q the quadrature axis as L L
APPENDIX
I:
SYNCHRONOUS
MACHINE MODEL
A..S. A.3
Various inductances in Che d and q axis. (Two axis model) In transient stability
studies,
small in magnitude.
(A.l)
algebraic, i.e.;
S t a to r e q u a t i o n s
"dt" dA.
+ V
r.i, ^0 dt
(A.8)
"
r J
Q Q
We now substitute the hybrid relations (A.5) and (A.6) in (A.8) for thc for thc variables A ^ , ip, iD and i^ to get variable; Stator equations vd m rid  (IJH00+^)it]  I^00q1Xq %
=
"riq
+ w{L
MD0+?d}id
+ ,,>L
MDOeF1?,F
^DO^D
(A. 9)
_x
_2
r U
F MD0*'F d "
FF
>*F
+ V
^W^S^D
F "IF * r n LMD0D X d
+ r L
E Mr)0^F
*D
dA
1.
"W^Q
.1
>X
,2 Q
APPENDIX I:
243
Define
To
do
= rQlQ5WQ2>These three time constants are respectively called the field, direct and quadrature axis open circuit time constants. l"
+ _
L
MDO +
ancl L
g'
= L
HQO
*"q ^ave keen previously defined as the direct axis and quadrature axis
Vq
= riq +
+ *%s^i%!
"
1
^ " ^WF^d
dA
V!S + V*DO V D
l
+ r L
l
W =
V D VN
M O
Q m
D MDOKF
" VT*>
(A 13)
d>, "dt
rA^K 1 !. ~ A _ / T '
" 0"MQO"Q **q
"q'
' qo
5 Eq =
"W ^F
F
(A 14
* >
^^0%
APPENDIX I:
243
Rotor equations
/T
vg,
/I
i^, /3 I (A.14) 
Substitution of
(A.16) in (A.12)
and the differential equations (A.18) for the rotor are obtained Stator equations
(A.1 7 )
V
q " " rI q
<fo
Uot.or equations
tor
dt T + ^d + ^D T 4
(x x t ) do (x d"V 'do
q ;
j_
do
(x
u ^x,
q
d x .'
"do
<1 E"dT $ 
"Ed V
Vq
i r.2 '3
(x
"if"* !
x
d x d'
7
1
x
(x
d  x J
d" x f: ''F
do
d x{) x,)
JL.
 T7> 
do
1 1
T
1
(x
"Ld' % x(  "V
uiL' '
(A.17) and (A.18) are the basic equations Ot the synchronous machine in the hybrid state variables i^, i^, E^, E^*, E^".
APPENDIX I:
245
Since we have neglected the stator transients, i.e. and dX ^^f Lhe differential equations of the stator reduce to
jt
at
algebraic equations. Electromechanical Rotor Equations To Eq. (A.17) and (A.18), we must add the mechanical equations The rotor angle is chosen relative to If 0 is
a synchronously rotating reference frame moving with constant angular velocity to^ electrical radians/per sec. measured with respect to a fixed reference, then U (ii t + a + elec. radians/sec. where a is constant, o The angle a  r,/2 if the d axis leads the q axis and equal to q axis leads the d axis. fll. as The mechanical
2ero
if the
We follow the former convention following Ref. equation of motion has been derived in Chapter III
. S% +
dt
i.0. where P
.
_ p11
5
d6
dt2
(A.19)
dt
m radian and
2
g
dt
+ D
ra
s
As shown in Ref. [1] after the p.u.
and P
is in radians.
Hd
q 
(A.S),
(A.14}
g=
(E
qVEdV + It^'Vq
d (A. 19}
(A 21
" >
becomes
H  IT  + %
Equations (A.17), (A.18), and disturbances. equations. (i)
s'Vq^a' 1
*:;*:?\Soj
d i fferentialalyebraic model of the synchronous machine and is found adequate for most multimachine simulation studios involvinq larqc There are five differential equations and two algebraic .''Several levels of simplification can be made i n the model We consider two such models (ii) the classical model.
depending on the needed accuracy of results. one axis Kt* model and
A. 4 One Axis Model q In this model there are no damper windings. Consequently x'V* x ., = X " " * x_, t  t "\ = 0 and 1. There are no d d' q q g do qo %
x''
%, 
APPENDIX I:
245
E\" and
d
E".
E'*'* = q q
and d
E''S  El
a
T.a
do
E' +
4 t ,
(E E)  Ifc ^do
where li = ^ (A,20) as (A.24)
p^ is derived independently from the expression P = E'l + (x'x')I,I y q q d q d q Hence the rotor equations are
a u: 5 _ dt *
The algebraic equations arc V.,.  ri,  x'l d d q V q ri g + E' q d + x'l. a d q + E'
(A.26 )
Suppose now we ignore the transient saliency so that x^j = x^. Then the algebraic equations V, a V <1 and the differentia] , (A.26) become ri   xpi + EC d d d d + x"l q + E'" d q q {A.23) and (A.26) become dE'
= rl
(A.27 )
equations
dt M d3 dt
B w
q q
* ft
 4r*
The two algebraic equations (A.27) can be converted to a phasor equation as follows: From Park's transformation with balanced conditions, thc voltage of phase a is (I]
V
N/
I (vdcost:'+vqsin(J>
the phasor representation is
(A.29)
Since 0
= u >^t
+ $ + t/2,
APPENDIX
1:
747
i s the angle between the network reference frame and the q a x i s of the (Fig. A.7). The network is assumed to be i n steady state. can be expressed either in the network In the latter case, thc
machines
jvd
The algebraic equations {A.27) can then be expressed in either of the two reference frames as V"t = rJt i E'  D'X^I Vfc = rlt + F.'  jx^it {Network Reference) (A.32) (Machine Reference)
248
The components ot the network reference frame phasors are the projections alonq the system D and Q axes and Lhe components of the machine reference frame phasors are along the machine d and q axes. equivalent circuit in the network shows the components in the two reference frames, The transformation between the two phasors is given by It =
X
Figure
(A.6 )
showsthe (A.7)
reference
tt&**
+
and
E
D'
t = *q
%<
q n
Notice thai. S
is the angle between the synchronously rotating network Q and 0 are the network reference axes in
and q and d are the machine reference axes and a phasor can be described The transformation of components between the two reference frames is easily verified as q d Q D cosa sin5 sino cos6 q d cos6 siri6 sin cos (A.33 ) g D
The transformation applies to both the current and voltage variables. A.5 Classical Model
To make the final transition to the classical model, we make the assumption that the field voltage Epn may not change very fast so that E^ remains more or less constant during the 01 sec interval of interest in transient stability studies. is true if T j i do This fairly large. We further assume that r=0 This implies and E' = E' i.e., 0. 1 that the q axis The phasor diagram now q coincides with K becomes as shown in Fig. A8. It is seen that y = & &o that phase angle of E' coincides with the rotor angle with respect to the
APPENDIX
I:
249
behind
direct
axis
transient
q AXIS
d AXIS
A,'/, model.
one axis
(SYSTEM REFERENCE)
APPENDIX
I:
249
i
A.8. model. Phasor diagram tor the classical
2b(J
synchronously rotating network reference frame. vice versa thus providing a big advantage.
transform the phasor quantities from network to machine reference frame and Hence we have the oft quoted assumption that the phase angle o f the voltage behind transient reactance coincides with the mechanical angle relative to the synchronously rotating reference frame. p  K'I g q q E' ! T t'cos<ofi+<to The expression for electrical power becomes
E* itl (eostSoos (fl<p) +sinSsin {&*))  Re{EI*} A_._6_ Flux Decay Model Used The model used in Sees. one given by H {A.23) P in Chapters III and IV
3.8 and 4.8 is actually a combination of the one The differential equation for E^ is the
d t
g (A.35)
 Pm  RelE'l*]
The
model is to avoid Lhe cumbersome network to machine transformation and viceversa. Thus while wc approximate the electrical power P^, the tlux decay effects approximate the electrical j arc modelled through References 1. (A.23).
Anderson, P. M. and Fouad, A. A., "Power System Control and Stability", (Book), Iowa State University Press, Ames, Iowa, 1977. Kiaz, M. , "llybridParameter Models of Synchronous Machines, IEEE Trans., Vol. PAS93, 1974, pp. 849858.
2.
APPENDIX
I:
251
Bien plus que des documents.
Découvrez tout ce que Scribd a à offrir, dont les livres et les livres audio des principaux éditeurs.
Annulez à tout moment.