Vous êtes sur la page 1sur 8

410

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 1, FEBRUARY 2004

Identification of Electromechanical Modes and Placement of PSSs Using Relative Gain Array
, Senior Member, IEEE, and Alfonso C. Serrano Duque, Student Member, IEEE Jovica V. Milanovic
AbstractThe paper investigates the possibility to use relative gain array (RGA) for power system small-disturbance stability analysis. The RGA has been successfully used in the past as steady state measure of interactions for multivariable, decentralized control in chemical and process engineering. The effectiveness of the RGA in identifying electromechanical modes and selecting the optimal location for placement of PSSs is compared with the conventional approach that uses transfer function residues, eigenvalues and participation factors. Two different multimachine power systems are used for the case studies. Limitations and advantages of the use of RGA in this purpose are clearly indicated in the paper. Index TermsDamping, electromechanical oscillations, relative gain array, stability.

I. INTRODUCTION

HE optimal placement of power system stabilizers (PSSs) in a power system in order to achieve the best possible damping of electromechanical modes has been a subject of a large number of research projects in the past. The most frequently used approach to this problem was to perform modal and sensitivity analysis of the linearized set of differential and algebraic equations of the power system model. Based on this analysis the best location for installing PSSs in the power system is determined [1][4]. Magnitudes of the residues associated with the critical electromechanical modes of the open-loop transfer function between the specified input and output are used as the indication of the most effective location for installing PSSs. The problem with this approach is that it may be very time consuming, as it requires calculation of eigenvectors, eigenvalues and residues of rather large systems that may have several hundreds of state variables. The RGA was introduced as a steady state measure of interactions for multivariable, decentralized control [5]. It was later extended to the frequency domain. For a multivariable system, the RGA is defined as the matrix of relative gains. Relative gain for an input and an output is the ratio of the uncontrolled gain (gain when all outputs are uncontrolled) and the controlled gain (gain when all the other outputs are controlled). Some of the important properties of the RGA are that it is independent on scaling, it depends only on the plant model and not on the

Manuscript received July 1, 2003. is with the Department of Electrical Engineering and ElecJ. V. Milanovic tronics, University of Manchester Institute of Science and Technology, Manchester M60 1QD, U.K. A. C. S. Duque is with the Empresarios Agrupados, Madrid 28015, Spain. Digital Object Identifier 10.1109/TPWRS.2003.821454

controller and that the RGA elements measure their own sensitivity to uncertainty [5][9]. So far RGA has been mainly used in chemical and process engineering areas [5][9] as: 1) Measure of coupling or interactions between variables (particularly at crossover frequency). 2) Tool to select pairings among outputs and inputs for diagonal control. 3) Tool for detecting the presence of the right-half-plane zeros. 4) Tool to assess the robustness of the system, and as a measure of sensitivity to the input, output or model uncertainty. As mentioned previously, all these tasks have been usually carried out using modal analysis techniques in power system analysis. Modal analysis requires the calculation of eigenvalues, eigenvectors, participation factors and residues of the system state space matrix. This matrix however, in the standard differential form does not preserve the sparsity of the differential-algebraic form, so sparse techniques cannot be applied [1]. Therefore, modal analysis can be very computationally demanding, and sometimes limits the size of the system to which it can be applied. Though, there are very powerful computational engines available for handling systems of up to 1000 states it would be an advantage to come up with a method that offers the same information about the system behavior by using less computationally demanding algorithms. As the RGA is calculated on the basis of transfer functions between selected outputs and inputs, not state equations, the calculation of eigenvalues is not required, so RGA can be an alternative or an addition to the modal analysis for systematic application in large power system studies. It measures two-way interactions between a determined input and output. The RGA peaks at a particular frequency indicate that the plant is difficult to control (or close to instability). (For example, for frequencies in the range from 0.2 to 2 Hz the RGA peaks identify electromechanical modes with small damping.) Furthermore, for a given input and output signal, a high value of the RGA element at a particular frequency indicates the interaction between the corresponding input and output. Hence, a damping controller with the input and the output associated with this RGA element could be used to improve the damping of the mode corresponding to this frequency. The main burden in the RGA calculation is the inversion of the matrix whose size is determined by the number of the inputs and outputs of interest. (For example, one input and one output per generating unit result in 100 100 matrix in a 100-generator system while the state space matrix of such system could be of the order 1500 1500!) The subject of this study therefore is the assessment of the RGA applicability in power system studies and comparison of the RGA, residues and participation factors approach in

0885-8950/04$20.00 2004 IEEE

AND DUQUE: IDENTIFICATION OF ELECTROMECHANICAL MODES AND PLACEMENT OF PSSs MILANOVIC

411

the study of electromechanical modes and optimal location of PSSs. The initial results of this study were reported in [10]. II. RELATIVE GAIN ARRAY For the completeness of discussion some basic definitions and methods for calculating RGA are given in this section. A. Definition of the RGA Let a multivariable square system be defined by the matrix , where . The relative gain, equation , for a given input and output is defined as the ratio between the uncontrolled gain and the controlled gain. The uncontrolled gain is the gain between the input and the output when all other outputs are uncontrolled (Fig. 1). This gain is the eleof multivariable transfer function ment
Fig. 1. Uncontrolled gain.

(1)
Fig. 2. Controlled gain.

The controlled gain is the gain between the input and the output when perfect control is exercised at the rest of the outputs [i.e., their value is zero (Fig. 2)]. Hence, for the output and the input , RGA element measures the effect that the control of the rest of the variables has on gain between and . B. Calculation of the RGA A direct or uncontrolled gain is given by the elements of the multivariable transfer function . A controlled gain is the corresponding element of the inverse transpose of the multivari[5]. Hence, the relative gain able transfer function, and output is: for input (2) And the RGA is the matrix of all relative gains (3) where is the element-by-element multiplication or Schur product. It is important to note that RGA is calculated on the bases of the multivariable frequency response. In frequency domain, it involves inversion of complex matrix whose size is determined by the number of the inputs and the outputs to the system. C. Main Properties of the RGA Some of the main algebraic and control properties of the RGA are: i) All rows and columns sum to one. ii) Any permutation of the rows and columns of the state matrix results in the same permutation in the RGA, i.e., the individual relative gains do not depend on the order in which inputs and outputs are considered. iii) Let G be a system with an isolated subprocess H. is equal to the elements of The RGA of the sub-process corresponding to the sub-process H. All other elements of rows and columns corresponding to the sub-process are zero. iv) The

RGA depends on plant model only and not on the controller. It is calculated for the open loop system. v) It is independent on the input and the output scaling, i.e., it is not influenced by the choice of units. vi) If one input (or an output) variable can be related to only one other variable (and some other variables which are not influenced by the rest of the inputs and the outputs) then the RGAs associated with these variables are the same [11], [12]. This relation may be algebraic or dynamic. (For example, the RGAs of a power system model when generator speeds or rotor angles are considered as the outputs are the same, because one variable is the integral of the other.) vii) A relative change in the element is equal to the negative inverse of its relative gain, to results in a sini.e., changing the element gularity of the matrix [13]. This particular property has two main implications: (1) A system having relative uncertainty of at the frequency larger the transfer function element than may have zeros at that frequency either on the imaginary axis or in the right half plane (RHP). (2) A relative change in the element of the state matrix greater than the inverse of the corresponding relative gain results in a singular matrix (i.e., the eigenvalue at the origin) [6]. viii) The negative inverse of the RGA element is equal to the ratio of relative change in the element of and its transpose inverse [14] (4) In case of large RGA elements, the elements of the inverted maare very sensitive to changes in the elements of . trix Therefore, plants having large RGA elements are very sensitive to model uncertainty [7]. ix) There is a relationship between the sign inversions of the RGA and the presence of RHP zeros [6]. is finite and different from zero. Assume that Consider a transfer matrix with stable elements and no zeros or . If and have different signs then poles at has a RHP at least one of the followings must be true: a) has a RHP transmission zero. c) (i. e. the zero. b)

412

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 1, FEBRUARY 2004

Fig. 4. Fig. 3. Block diagram of the test system. TABLE I ELECTROMECHANICAL OSCILLATIONS FOR TS1

Maximum RGA magnitude of each oscillation versus damping factor.

subsystem with input j and output i removed) has a RHP transmission zero. This property indicates that under certain conditions, the change of the sign of the RGA element between the steady state and high frequency is a sufficient condition for the existence of the RHP zero. III. TEST SYSTEMS Two power systems are used in the study described in the paper. A block diagram of the Test System 1 (TS1) is shown in Fig. 3. It consists of two areas. The Area 1 (A1) with generators 1 and 2 and load 4, and the Area 2 (A2) with generators 3 and 4 and load 14. The generator 5 and the load 21 are connected to both areas though, electrically closer to the Area 1. There is a power transfer from A1 to A2 of about 150 MW. All generators are modeled with detailed sixth order model and equipped with the AVRs. Loads are modeled as a constant impedance loads, except load 21. This load has a 40% component of induction motors. A full-load and a half-load operating condition of the system are analyzed. The second power system (TS2) used was New England test system with 10 generators and 39 buses. The data for the TS2 can be found in [15] or in the Matlab Power System Toolbox. IV. CASE STUDIES Firstly, the modal analysis of the system is performed and the electromechanical modes identified. Participation factors, right eigenvectors and residues of the open loop TF between the AVR reference voltage and the electrical power and speed are calculated. The results are summarized in Table I. It can be seen that there is an unstable electromechanical mode (0.98 Hz) associated with generators 1, 3 and 4. The rest of the electromechanical modes are stable but with a very low damping, A. Identifying Electromechanical Modes Following the modal analysis the RGA is calculated using (3), and setting voltage reference of the AVR as the input and

the generator speed and electrical power as the output. (This was done in two separate calculations in order to compare the results obtained with different output signals). The results of calculation obtained with the generator speed as the output are shown in Fig. A1 in the Appendix. It can be seen that the RGA has peaks corresponding to frequency of the electromechanical oscillations. Note that these peaks are higher for the generator states that participate more in particular mode. (For example, the subfigure at location 4-3 in Fig. A1 represents the RGA element calculated between the reference voltage of the AVR of the generator 4 and the speed of the generator 3.) RGA at a frequency For example, there are peaks for of 0.98 Hz, which corresponds to the first electromechanical mode. There are peaks in the diagonal elements corresponding to generators 1, 3 and 4 which, according to participation factors, are active in that mode. No peaks appear for the rest of the generators, which also have low participation factors (as confirmed by modal analysis). Furthermore, the highest peak corresponds to the generator with the highest participation factor. The off-diagonal elements at the RGA matrix show the interaction of the generators at particular frequency, as only the elements corresponding to generators which participate in that mode have appreciable peaks (See again subfigure 4-3 which indicates involvement of generators 3 and 4 in the identified modes.) The electromechanical oscillations having similar frequency complicate this analysis, because the peaks in the RGA may overlap. (Note that the resulting RGA matrix is symmetric, thus indion the output is cating that the influence of the input similar to the influence of on .) At higher frequencies, RGA tends to identity matrix, indiof cating that there is a strong coupling between and the same generator, and no interaction between different machines. At low frequencies however, RGA rises to a high value, indicating that this system is difficult to control in the steady state. This is not surprising, as is not an appropriate variable to control generator speed. It can be also seen that RGA peaks are higher for the modes closer to the imaginary axes, regardless of the sign of the real part. The relationship between these peaks and the closeness to the imaginary axis (measured in terms of damping factor) can be seen by examining damping factors and RGA peaks given in Table I and also from Fig. 4. For each mode, the highest RGA peak of all generators has been taken, and plotted against the corresponding damping factor in Fig. 4. It can be seen that the highest RGA corresponds to the mode closest to the imaginary axis and it decreases with the damping factor, with a single exception.

AND DUQUE: IDENTIFICATION OF ELECTROMECHANICAL MODES AND PLACEMENT OF PSSs MILANOVIC

413

TABLE II ELECTROMECHANICAL OSCILLATIONS FOR TS2

TABLE IV IDENTIFYING PARTICIPATING GENERATORS FOR TS2

TABLE III IDENTIFYING PARTICIPATING GENERATORS FOR TS1

Similar analysis was performed with a second test system, TS2. The results of this analysis are presented in Table II and Fig. A2 in the Appendix. This figure shows only diagonal elements (between the input to the exciter and the speed of the ) of the complete 10 10 matrix generator , representing the RGA elements between the input to the AVRs and the speed of the corresponding generators. The RGAs again have peaks that correspond to the critical eigenvalues. The peak magnitude is related to the closeness of the eigenvalue to the imaginary axis as before (observe the damping factors and the RGA peaks given in Table II). (For example, the RGA peaks corresponding to the frequency of the mode 2, which is far away to the right from the imaginary axis, are almost negligible.) The off-diagonal elements of the RGA correctly identify the groups of generators involved in each mode. (That is, the RGA corresponding to generators involved in particular modes have peaks at the mode frequency.) B. Comparison of the RGA and Modal Analysis The involvement of the generators of TS1 in different electromechanical modes is ranked in Table III using different indices. The first number in the relevant column indicates the most involved generator. Generally, there is an agreement between the participation factors and the speed and power residues based ranking. The exception is mode 1, where each one of the three gives a different ranking. (The power residues can be obtained by multiplying the speed residues with the corresponding inertia and a constant [7], so they are not expected to agree completely when the generators have different inertias as in this case.) The RGA (column 5 in the above table) provides almost identical information about the generator involvement as participation factors and residues. They all have the highest values for the same generators. In the case of the mode 3, however, the relative order of the two most active machines in the mode is reversed

when considering the RGA. The results above showed the relationship between the RGA and the classical modal analysis for diagonal couplings. The same analysis is performed with TS2 and the results are summarized in Table IV. It was found that the frequencies of the three fastest modes of the system ( 1.5 Hz) are very close. This makes it impossible to attribute the peaks of the RGA to one frequency or the other. Therefore, the comparison of the modal analysis indexes with the RGA for these modes is omitted. For the lowest frequency oscillation on the other hand, each index gives a different ranking of the activity of the generators. It should be noted though, that the same generators are identified as important by the speed residues and the RGA (this was also the case with TS1). (Note: the big differences in the ranking obtained using the speed and the power residues can be attributed to the fact that the generator 2 (representing the interconnected USA system) has much bigger inertia than the rest of the machines As the power residues are approximately speed residues, scaled by the size of the machine [7], this bigger generator will tend to come up as first when magnitudes of power residues are used for ranking.) After inspecting results shown in Table IV and comparing actual magnitudes of the RGA peaks it can be concluded that highest RGA magnitude peaks identify electromechanical modes close to the imaginary axis. If the system is stable, then the highest RGA peaks identify the most critical modes. If the system is unstable and its unstable poles are close to the imaginary axis the RGA peaks still give good results. The peaks of RGA, however, do not give information about unstable modes with large real parts. From Figs. A1 and A2 and Tables II and IV the following observations can be made: 1) The power and the speed RGAs have peaks that corresponds to the frequencies of the electromechanical modes. Their values, in diagonal and off-diagonal elements indicate which generators are involved in each mode. By comparing the peaks of the RGA and the results of the modal analysis the same conclusions about generator involvement in particular mode can be reached. 2) The electromechanical modes having similar frequency complicate the analysis, i.e., they cannot be effectively separated using the RGA analysis. 3) The largest RGA magnitudes are generally associated with the modes having lowest damping factor irrespectively whether they are in the left or the in the right hand side of the complex plane (i.e., stable or unstable). 4) The RGAs corresponding

414

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 1, FEBRUARY 2004

TABLE V PSS DESIGN PARAMETERS

TABLE VI EIGENVALUES OF TS1 WITH TUNED PSSS (REAL PARTS OF THE EIGENVALUES ARE IN 1/s AND IMAGINARY PARTS IN HERTZ.)

Fig. 5. Generator electrical power responses (TS1) following three-phase fault at bus 25.

to the speed and the electrical power output of the generator are very similar. Being almost equal implies that the RGA is not a suitable tool for selecting the most appropriate input to a damping controller. 5) The highest peaks of the RGA elements coincide with the highest residue magnitudes in most cases for both test systems. This makes the RGA suitable for detection of the best location for placement of damping controllers. C. Stabilizing the System Using the RGA In order to establish whether the system (TS1 and TS2) can be stabilized using only the information provided by the RGA analysis the following procedure was used. The open loop transfer function residue based method discussed in [1], [16], [17] is used to design the PSS though the other methods (e.g., frequency response) could be also used without any loss of generality of the discussion presented in the sequel. The method used is summarized here for the completeness of discussion. The relationship between the residue and the corresponding shift in the system eigenvalue is given by: (5) Where is the shift in critical eigenvalue after the clois the residue of the open loop sure of the feedback loop, and is transfer function corresponding to eigenvalue the transfer function of the controller in a feedback loop. (Note: If during the derivation of (5), a negative feedback convention is assumed there would be a minus sign on the left hand side of the expression (5). This would indicate that the shift of the closed loop eigenvalue would be in direction opposite to the phase of the residue. Alternatively, with a positive feedback assumption the sign would be plus and the eigenvalue shift would be in phase with the residue.) corresponding to the critical eigenvalue , The residue -th output and -th input of the open loop system transfer function (TF) is given by (6)
Fig. 6. Generator electrical power responses (TS1) following three-phase fault at bus 25 with a PSS at generator 3.

Fig. 7. Root loci of the eigenvalues of TS2 with PSSs at generators 4 and 6.

where and are the right and left eigenvector respectively, and are corresponding row and column of the output and and input matrix respectively. In order to achieve the best damping the critical eigenvalue should be shifted toward the LHP without any change in frequency. To achieve this, the shift of the critical eigenvalue must be in the direction in the complex plane. Hence, the phase compensation required (to be provided by the

AND DUQUE: IDENTIFICATION OF ELECTROMECHANICAL MODES AND PLACEMENT OF PSSs MILANOVIC

415

Fig. A1.

RGA for TS1 with voltage reference as the input and the generator speed as the output (vertical axis represents RGA magnitudes in per unit).

PSS in a positive feedback loop) can be calculated as the difference between and the residue phase angle (7) After calculating the phase compensation that needs to be provided by the PSS lead/lag block(s) the parameters of the PSS transfer function (8) (where N is the number of the lead-lag block, and are the is constants to be determined during the tuning process, the washout (high-pass filter) whose time constant is preset to typically 3 s10 s) are calculated as follows: (9)

(10) (11) (Note: Formula (10) is a general formula for either lag or lead compensator. In the case when a phase leg compensator is needed to stabilize the system, should be substituted as a negative.) First, PSSs were tuned for the TS1. For the unstable 0.98-Hz mode the RGA indicate generator 1 as the most efficient location

for installing a PSS. Speed residue however, points out generator 4 as the best location for the PSS. On the other hand, by far the largest RGA (20.92) and speed and power residues point out generators 3 and 4 as the best location for damping of 1.66 Hz mode (which has the lowest damping ratio). Therefore PSS is firstly added only to the generator 1 and tuned with respect to 0.98 Hz mode, then only to generator 4 and tuned with respect to 1.66-Hz mode and finally only to generator 3 and tuned with respect to 1.66-Hz mode. The designed PSS parameters are given in Table V and the eigenvalues obtained in all cases are listed in Table VI. Generator responses following a three phase fault at bus 25 lasting 100 ms in the system without PSS and with a PSS at generator 3 are shown in Figs. 5 and 6, respectively. Fig. 6 illustrates successful stabilization of the system with a single PSS using the RGA information only for the placement of the PSS. It can be seen from Table VI that the PSS at any of the three identified generators stabilizes the system and that by installing the PSS at the generator corresponding to high RGA value the system can be successfully stabilized. By inspecting the magnitudes of the RGA peaks listed in Table IV for TS2 it can be seen that the highest peak (at a frequency of 0.64 Hz) of the RGA corresponds to the generator 6. So the PSS was placed at this generator. The PSS stabilizes initially unstable 0.64 Hz mode however, the damping of the mode is still very low and the PSS had little effect on the rest of the unstable modes. (Note that the residues also indicate a low sensitivity of these modes to a feedback at generator 6). The RGA was subsequently recalculated for the new system with a PSS.

416

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 1, FEBRUARY 2004

Fig. A2. Diagonal elements of the RGA for TS2 with voltage reference as the input and the generator speed as the output.

The peaks of the RGA corresponding to the 0.64-Hz mode are now lowered. The highest RGA peak for the new system occurs at the frequency of the mode 3 (1.07 Hz), and corresponds to generator 4. After tuning a PSS at generator 4, mode 3 is shifted further to the left (see Fig. 7). The damping of this mode is limited by the influence of a close zero. The second initially un-

stable mode (0.98 Hz), however, remained unstable. Following the same procedure the next PSS should be placed on generator 3 with the strongest influence on 1.26-Hz mode. The unstable 0.98-Hz mode however, is very little sensitive to a feedback at this location and it would remain unstable. This analysis confirms previous conclusion that the RGA can not identify modes

AND DUQUE: IDENTIFICATION OF ELECTROMECHANICAL MODES AND PLACEMENT OF PSSs MILANOVIC

417

(eigenvalues) with large positive part. Fig. 7 shows root loci of the electromechanical modes of TS2 with PSSs at generators 6 and 4. The x signs mark the original location of the pole, signs mark the tuned value and o signs mark the location of corresponding zeros. V. ADDITIONAL ANALYSIS In order to fully explore the possibility to use the RGA in power system stability studies several other properties of the RGA listed in Section I are investigated. Due to the space limitation only few major findings are listed here. A. Interaction Between Control Loops During the study of the interactions between voltage and frequency control loops, it was found that the RGA clearly identifies these interactions at the frequency of the electromechanical modes. Therefore, in accordance to [9], the inverse-based controllers should not be used in voltage and power control loops of the generators significantly involved in the electromechanical modes. This is especially true for the generators where the PSS is to be installed. B. Relation Between RGA and Right Half Plane (RHP) Zeros A theorem given in [6] represents a tool for identifying the RHP transmission zeros and transfer function zeros, through their relationship to the change of sign of the RGA gains between the steady state and high frequencies. This is an important issue as both RHP transmission and transfer function zeros are known to be problematic for the control [17]. The validity of this theorem was tested on a single-machine system and it was found that a sign inversion of the RGA is associated with a RHP plane zero at a corresponding frequency. This method however, does not guarantee the absence of the RHP zeros, as it provides only a sufficient condition for their existence. C. Robustness to Change in System State Matrix Elements The RGA was found useful in identifying the elements of the system state-space matrix whose uncertainty in may yield system instability. The RGA approach was compared with participation factors as a measure of sensitivity to the change in matrix element. The RGA measure however, is only applicable when the instability is associated with a single real pole crossing the imaginary axis. It is not valid for the case when instability is caused by complex modes (which is often the case in power system studies). It is also important to note that these results are related to individual matrix element uncertainty. The elements of the matrix of the realistic power system however, are not independent. VI. CONCLUSION The paper presented exploratory analysis of the possibility to use the relative gain array for identification of electromechanical modes and placement of PSSs. It was found that the peaks of RGA magnitudes correctly identify the frequencies of the poorly damped electromechanical modes. The RGA peaks magnitudes agree with participation factors in determining which generators are involved in particular oscillation. They also agree with residues when determining

potential locations for placing PSSs and such can be used for placement of PSSs. Some identified drawbacks of using the RGA are that they do not indicate whether poorly damped oscillations are stable or unstable. In particular, RGA does not identify modes with a large positive real part and additional (e.g., modal) analysis is needed in such cases. Being a frequency response method, RGA is difficult to apply when analyzing systems having electromechanical modes with similar frequency, because the RGA peaks may overlap. The other known properties of the RGA have also been tested and their applicability to power system analysis discussed. REFERENCES
[1] P. Kundur, Power System Stability and Control. New York: McGrawHill, 1993. [2] M. Klein, G. J. Rogers, and P. Kundur, A fundamental study of inter-area oscillations in power systems, IEEE Trans. Power Syst., vol. 6, pp. 914921, Aug. 1991. [3] F. L. Pagola, I. J. Prez-Arriaga, and G. C. Verghese, On sensitivities, residues and participations, applications to oscillatory stability analysis and control, IEEE Trans. Power Syst., vol. 4, pp. 278275, Feb. 1989. [4] I. J. Prez-Arriaga, G. C. Verghese, and F. C. Schweppe, Selective modal analysis with applications to electric power systems. Part I, heuristic introduction. Part II, the dynamic stability problem, IEEE Trans. Power App. Syst., vol. PAS-101, pp. 31173134, Sept. 1982. [5] E. H. Bristol, On a new measure of interaction for multivariable process control, IEEE Trans. Automat. Contr., vol. AC-11, pp. 133134, Jan. 1966. [6] M. Hovd and S. Skogestad, Simple frequency-dependent tools for control system analysis, structure selection and design, Automatica, vol. 28, no. 5, pp. 989996, 1992. [7] S. Skogestad and M. Morari, Implications of large RGA elements on control performance, Ind. Eng. Chem. Res., vol. 26, pp. 23232330, 1987. [8] S. Skogestad and K. Havre, The use of RGA and condition number as robustness measures, Comput. Chem. Eng., vol. 20, pp. S1005S1010, 1996. [9] K. Uhlen, B. A. Foss, and O. B. Gjsterw, Robust control and analysis of a wind-diesel hybrid power plant, IEEE Trans. Energy Conversion, vol. 9, pp. 701708, Dec. 1994. [10] J. V. Milanovic and A. S. Duque, The use of relative gain array for optimal placement of PSSs, in CD ROM of the IEEE Power Eng. Soc. Winter Meeting, Columbus, OH, Jan. 28Feb. 1 2001. [11] F. G. Shinskey, Distillation Control: For Productivity and Energy Conservation, 2nd ed. Westport, CT: Greenwood Press, 1984. , Process Control Systems: Application, Design and Adjustment, [12] 3rd ed. London: McGraw-Hill, 1988. [13] C. C. Yu and W. L. Luyben, Robustness with respect to integral controllability, Ind. Eng. Chem. Res., vol. 26, pp. 10431045, 1987. [14] Z. X. Zhu, Variable pairing selection based on individual and overall interaction measures, Ind. Eng. Chem. Res., vol. 35, pp. 40914099, 1996. [15] K. R. Padyar, Power System Dynamics Stability and Control: J. Wiley, 1996. [16] D. R. Ostojic, Stabilization of multi modal electromechanical oscillations by coordinated application of power system stabilizers, IEEE Trans. Power Syst., vol. 6, pp. 14391445, Nov. 1991. [17] L. E. Jones and G. Andersson, Application of modal analysis of zeroes to power systems control and stability, Elect. Power Energy Syst. , vol. 46, pp. 205211, 1998. (M95SM98) received the Dipl.Ing. and M.Sc. degrees Jovica V. Milanovic from the University of Belgrade, Yugoslavia, and the Ph.D. degree from the University of Newcastle, Australia. Currently, he is a Reader in the Department of Electrical Engineering and Electronics at the University of Manchester Institute of Science and Technology, Manchester, U.K. Alfonso C. Serrano Duque (S00) received the Ingeniero Industrial degree from the Universidad Pontificia de Comillas, Madrid, Spain, and the M.Sc. degree from the University of Manchester Institute of Science and Technology, Manchester, U.K. Currently he is an Electrical Engineer with Empresarios Agrupados, Madrid, Spain.

Vous aimerez peut-être aussi