The 20th Iranian Electrical Power Distribution Conference (EPDC2015), 28-29 April 2015, Zahedan, Iran

Distributed Predictive Control for power balancing in

electric distribution networks
Hamidreza Iranmanesh

Ahmad Afshar

Department of Electrical Engineering organization

Amirkabir university of Technology
Tehran, IRAN
iranreza7@aut. ac. ir

Department of Electrical Engineering organization

Amirkabir university of Technology
Tehran, IRAN
aafshar@aut.ac. ir

Abstract- One of the efficient control method is the model

predictive control

(MPC)

that calculates

The case study is a power grid with 4 control zones that

each of which has local generation and load. The control
method used in this study is introduced in [8] which is a
different method for distributed predictive control based on
cooperation between controllers and can guarantee system
stability.

appropriate inputs

corresponding to desired output by an optimization method and

predicting future output based on related model. Then the first

step of calculated input is applied to system. AN important

application

of

this

control

method

is

for

controlling

load

frequency and power balancing in power grids. It preserves

transition power AND network frequency as one of important

parameters

of

power

quality

within

admissible

range

Among the implemented innovations in this paper is that

the load changes are applied to the system in a pulse shape
rather than stepwise and with much delay to examine the
system perfonnance and speed to return to equilibrium mode
after successive load changes. Also, load disturbance is
increased up to 20% and applied to the system IN a new fonn so
that in zone 3 it appears as a load increase and in zone 2 it
appears as a load decrease. In addition, the range of control
input changes is decreased by 60% to avoid saturation. In other
words , the constraints of system are chosen tougher to
examine system perfonnance more accurately in the new
conditions. Finally, in order to evaluate overall performance of
the system, total error is plotted. It is a weighted combination
of frequency changes and transmission power changes
regarding its equation.

by

controlling generation. Considering the complicity of calculations

in centralized architecture in large scale systems, in this paper a
cooperative distributed predictive control is used that makes the
calculations faster and easier and also offers good performance.

HOWEVER in the not too distant future, with development of

smart grids , Exploitation of distributed generation in demand
side and electric distribution network will be usual and it will be
necessary to replace traditional control methods by smart and
advanced methods.

In this regard, results of the DISCUSSED

control method are compared with those of the classical control

of load-frequency and its proper performance is shown against

load variation disturbance, with considering a power grid with 4

control zones that each zone has local generation and load.

Keywords- power balancing; smart grid; predictive control;

frequency control

II.
I.

INTRODUCTION

In the MPC strategy, a model is used for predicting states

and future outputs of the system over a fmite horizon called the
prediction horizon. Then, regarding the difference between
predicted output and the desired output; an optimization
problem is solved to minimize the related error. The control
input is calculated over a fmite horizon called the control
horizon but only the first step of calculated input is applied to
the system at each sampling time. At the next sampling time,
this process is repeated by moving the horizon forward. So, this
type of control method is termed as receding horizon or
moving horizon method. For more infonnation one can refer to
[5] .

This template, Considering the various types of equipment

and subsystems used in power network and also electric
distribution network , such networks are considered as large
scale systems. In this regard, more detailed study is available in
[1] . Centralized control of such systems with regard to the
complexity of calculations and the large number of variables is
a difficult and impractical task. An appropriate solution is to
solve the control problem through decentralized or distributed
architecture by separating the overall system into smaller
subsystems [2, 3 ] . Currently one of effective control methods is
the MPC that is appropriate for multi variable constrained
systems and industrial applications. A comparison of several
control methods for controlling the frequency of a power grid
is done in [4] . In this regard, in the present paper, changes in
system characteristics and related parameters and or their
completion are used to examine control methods in new
conditions more precisely and completely.

978-1-4673-6612-0115/$31. 00 2015 IEEE

Downloaded from http://www.elearnica.ir

MODEL PREDICTIVE CONTROL

A.

Predictive control in large scale structure

As mentioned in the introduction, due to computational

complexities in centralized predictive control method in large
scale systems, despite relatively good performance of this
method, in many cases it is not practical. Therefore, separation

212

of an overall system into various subsystems produces a more

simple structure for calculations so that each subsystem has a
local controller. One of these structures is known as
decentralized MPC. In this case, it is assumed that no
interaction exists between the local controllers. In this
architecture, due to ignoring some communications, system
performance may be degraded and this can also have a negative
effect on system stability. This type of control can be
acceptable for systems with poor communication between their
subsystems. Some aspects of this control method have been
studied in [6] .

s.t.

Xj(k+qj + I) AjjX jj(k+qj)+Bjjup+q)+Bdjjdj(k+qj)+

=

I AjiXi (k+q)+BjiUi (k+q)+Bdii (k+q)

;t

Where mE Z+ indicates the number of iterations in the

iterative

But other type of large scale structures has a distributed

architecture that has neither the computational complexity of a
centralized architecture nor the performance disadvantage of a
decentralized architecture. For this reason, a type of this
architecture has been used in this paper that its specifications
will be described in the next sections.
B.

Np

Where AE R nxn ,

BE R nxm

coefficients and x ERn

, U ERm

, dERS

{x
. min
[IIj (k), ,u/k+Nc-1)]

...

mT

(k+NI')Px (k+NI')+

time

for

Ne 1, Ne

are weight

set).
E

Z+

is

symmetric

x/k) is considered as the initial

beginning

the

prediction

and

(k) E

ER

n/sj

are

matrix

Sj

(3)
Also, there are the following equations:
M

s] =s, W >- 0, I w] = l,x = [x; ... x:f (4)

In
J = m, I
i =n, Im
i j=1
j= 1
j=1
j=1

by
are equation matrix

In addition, the feasible input sequences are characterized

set and the following equation:

Uj

(5)

are vectors related to

Terminal penalty matrix P constrains terminal state

x(k + Np) to terminal set s N (rN ) and therefore, a priori
P

Distributed optimization problem

R . >- 0
}

'

integer

coefficients and
R
is the disturbance input of j-th
subsystem at time k with respect to the condition (3).

guarantee for system stability is provided[4] , so that there are

the following equations:

J-th subsystem control law in cooperative distributed

predictive control structure is obtained by solving the following
optimization problem :

<l>j'
j

at

non-negative

is prediction horizon and

horizon. Q. >- 0
J

dj

states of the system, control inputs and bounded disturbance

respectively.
A.

is

(1)

Bd E R nxs

(z+

q; E {O, 1,2,...,Np -I}.

SYSTEM SPECIFICATIONS

Ax(k) + Bu(k) + Bcfl (k)

Z+ ,Np I

state

A system is considered consisting of M subsystems that in

general, its state space equation is as follows:

k + 1)

algorithm

matrices. Also the value of

Considering that ignoring interactions between local

controllers in a decentralized structure (especially in cases
which there are strong interactions) can have a significant
effect on system perfonnance degradation, therefore as a
solution, a distributed architecture can be used where different
controllers communicate with each other. A type of distributed
structure that is introduced and used in [8,9] is known as
feasible cooperation based MPC (FC-MPC). In this framework,
a global optimization problem is solved cooperatively to
compute each control input and stability of the equilibrium
point is guaranteed with regard to related conditions. However
this situation requires more communication rather than a
decentralized structure.

x(

control

Distributed MPC(DiMPC)

III.

jE{I,2, ...,M}

i =l , i j

(6)

(7)

(2)

Where Sq (rq) is a closed set and a subset of Rn . Matrix P

N
T
N -I
T
L= w}Lt=f[xj (K+t)Q/j(K+t)I+Lt= [uj (k+t)Rjuj(k+t)])}
JI

and K are obtained by solving an unconstrained infinite

horizon LQR problem as follows [10] :

213

P=(A+BK)P(A+BK)+KRK +Q

(8)

K=-(R+BPB)BPA
If K= 0 then (8) reduces to the Lyapunov equation

10].

IV.

[4,

In each zone, local load variation causes variation in

frequency. MPC controller drives the frequency variations and
the transmission power variations in tie-lines to zero by
adjusting PR control input. In this case, load demand change is
considered as external disturbance.

7,

ApPLICATION TN POWER GRID

Specifications of simulated system

A.

The problem of load-frequency control in power grids can

be solved in a good way through predictive control [4, 9, 11,
12]. Considered power grid that can be an electric distribution
network with small scale generators as a part of a smart grid, is
composed of 4 control zones and tie-lines that provide
connections between the zones. Each zone can contain several
generators and several consumers that all of them can be
modeled and integrated as a generator and a consumer.

It is assumed for k:,,:: 20 the grid is in the steady state

t>.mj

and

= M'lf = M'Wj = 0

disturbance
M'w -OJ , M'
L
2

For

20 < k:,,:: 70

the external

in
the
value
of
enters the system in the form of pulse in

-tDJ

zones 2 and 3 . Also Np=60 , Nc=1 and simulation time is 300

and the maximum number of iterations of the algorithm at each
time sample is mmax= 8. In addition, the range of control input
changes is determined as follows:

Regarding the mentioned references, since automatic

generation control is limited to relatively small disturbances, so
the dynamics of each zone can be linearized. Regarding to the
references mentioned at the beginning of this section, the
standard model for zone j is obtained and the next calculations
are done as follows:

-0.2 M>R
J

+0.2 , j E {1,2,3,4}

(14)

State variables are defined as follows:

(9)

Zone 1: MVAI

(10)

MMEI

.:1 WI

Zone 2 : .:18'2

MVA2

MME2

.:1W2

(11)

Zone 3

.:1823

MVA3

MME3

.:1W3

(12)

Zone 4 : .:1834

MVA4

MME4

.:1W4

B.

(13)

Automatic generation control in classical method

The traditional automatic generation control(AGC) method

uses the PI controller. In this regard, the following controller is
used for zone j [4]:

In this modeling, the symbol L1 indicates deviation from

steady state and NEj represents the number of neighbors of
subsystem j. This power grid is shown in Figure 1. Also, the
list of used parameters is introduced in table 1.

MR =- Rj
J
Rj

Ej

1'=-00

NE
j
(E jL1OJj+

L Mh )dt'
h=1

(15)

are PI controller tuning parameters.

Error is calculated for each zone by the following equation

Figure 1.

[13]:

A power grid consisting of 4 control zones

214

TABLE 1- THE PARAMETERS USED IN THE SIMULATION

kl
PT1E

tie line power flow between areas k&l

OJ

angular momentum

phase angle

governor time cons tan t

PME

mechanical power

PVA

steam valve position

to

angular frequency of system

(%change in load) / (%change in frequency)

F
r

(%change in frequency) / (%change in unit output)

Pw

nonfrequency sensitive load

SkI

tie line stiffness coefficient ( between areas k &1)

PR

load refrence setpoint

tc

charching time cons tan t

ERRORik)

hj

NEj

..

I'1OJik) + L I'1PjJE(k)
i=I,i*j

(16)

bj is the relative weighting coefficient of error outputs.

Total error is determined by the following equation:

I. =IERRORj(k)
(17)
In order to examine and compare overall performance of
the system, the outputs related to zone I (that is without
disturbance input) and zone 3 (that includes disturbance input
as load pulse change) and the total error of the system are
shown in Figure 2. Total error is collective combination of
existing errors including frequency variations and transmission
power variations between the zones.
TOTAL_ERROR(k)

(b)

(c)
Figure 2. The results related to classical AGe : a) frequency changes in
zone I , b) frequency changes in zone 3 , c) total error

C.

Generation control by the FC-MPC method

In Figure 3, the same as previous section, to evaluate

overall performance of control system, curves of frequency
changes for zone 1 (without local load changes) and zone 3
(with local load pulse changes as external disturbance input)
and curve of total error are shown. Figure 3 clearly
demonstrates the error caused by load disturbance that includes
the deviations in frequency and transmission power, can be
gradually decreased by good control performance.

(a)

215

with the expansion of the smart grids, the use of distributed

generators will expand on demand side and in the electric
distribution networks. Hence, it seems essential that the
modern control methods with advanced specifications should
be used in place of traditional methods. In this regard,
predictive control can be used as a efficient control strategy
that offers a kind of foresight in design and implementation of
a control system.
According to the plotted curves, it may be concluded that
using a cooperative distributed control structure can fulfill
overall goals of the system regarding to cooperation and
interaction between controllers of different zones, in addition to
facilitating calculations in comparison with centralized control
structure. However, with increasing communications, the
system complexity is increased somewhat rather than the
decentralized structure and computational speed IS DECREASED,
but it should be noted that AGC is a relatively slow and precise
process [14] and increasing cooperation between control zones
can improve system performance.

(a)

Of course in the new researches to improve this structure,

techniques are still being studied and offered.
ACKNOWLEDGMENT

We are grateful of Kerman Regional Electric

Company(KREC) cooperation in relation to the discussed
subject.

(b)

REFERENCES

[1]

Dragoslav D. siljak (1991). "decentralized control of complex systems"

Academic press limited.

[2]

N. R. Sandell, Jr., P. Varaiya, M. Athans, and M. Safonov,(1978)

"Survey of decentralized control methods for large scale systems" ,IEEE
Trans. Autom. Control, vol. 23, no. 2, pp. 108-128,

[3]

Dragoslav D. siljak, (1978)" Large-scale Dynamic Systems: Stability

and Structure" . North-Holland.

[4]

Ralph M. Hermans, Andrej Jokic', Mircea Lazar, Alessandro Alessio,

Paul PJ. van den Bosch,Tan AHiskensand Alberto Bemporad ,(2012)
"Assessment of non-centralized model predictive control techniques for
electrical power networks" , International Journal of Control ,Vol. 85,
No. 8, 1162-1177.

[5]

Liuping Wang ,(2009) "Model Predictive Control System Design and

implementation using MATLAB" , Springer.

(C )
Figure 3.

The results related to FC-MPC : a) frequency changes in zone

frequency changes in zone 3 , c) total error

I ,b)

[6]

Figures indicate that control of frequency in considered grid

is done properly by cooperative distributed predictive control
even with successive load variation. Finally after a certain time
of load disturbance, the grid returns again to steady state by
decaying frequency deviations and created error.
V.

CONCLUSION

One of important subjects for presserving power quality in

power networks is the appropriate control of frequency within
admissible range, especially after making them smart in future.

Cui, H., & Jacobsen, E. (2002), "Performance limitations

decentralized control", Journal of Process Control, 12,485-494.

in

[7]

Alessio, A, and Bemporad, A ,(2007), "Decentralized Model

Predictive Control of Constrained Linear Systems", in Proceedings
European Control Conference, Kos,Greece, pp. 2813-2818.

[8]

Venkat, AN. ,(2006), "Distributed Model Predictive Control : Theory

and Applications" , PhD Thesis, University of Wisconsin-Madison,
Madison, WI, USA

[9]

Venkat, AN., Hiskens, l.A, Rawlings, J.B., and Wright, S.J. ,(2008),
"Distributed MPC Strategies with
Application to Power System
Automatic Generation Control" , IEEE Transactions on Control
Systems Technology, 16, 1192-12

[10] Mayne, D.Q., Rawlings, lB., Rao, C.v., and Scokaert,P.O.M. ,(2000),
"Constrained Model Predictive Control : Stability and Optimality",
Automatica, 36, 789-814.

It would not be unexpected that in the not too distant future

with activity of power quality market , this issue will also have
specific and measurable economic value. On the other hand ,

216

[II] Kundur, P.,( 1994), "Power System Stability and Control" ,New York,
NY, USA: McGraw-Hill.
[12] A. J. Wood and B. F. wollenberg ,(\996), "Power Generation Operation
and Control" , New York: Wiley.
[13] Miaomiao Ma, Hong Chen , Xiangjie Liu , Frank Allgower , (2014) ,
"Distributed model predictive load frequency control of multi-area
interconnected power system" , Electrical Power and Energy Systems 62
, 289-298.
[14] Jaleeli, N., VanSlyck, L., Ewart, D., Fink, L., and Hoffmann, A. ,(1992),
"Understanding Automatic Generation Control" , IEEE Transactions on
Power Systems , Vol. 7, No. 3" 1106-1122.

217