Second IEEE Conference on Control Applications, September 13 - 16, 1993 Vancouver, B.C.

DISCRETE-TIME ROBUST CONTROLLER FOR LOAD FREQUENCY CONTROL OF POWER SYSTEMS Youyi Wang Rujing Zhou Wei Kai Chua Felix S. M. Fong School of Electrical and Electronic Engineering Nanyang Technological University Singapore 2263

ABSTRACT In this paper, a discrete-time version of a robust controller (see (3]), based on the standard Riccati equation approach, is proposed for power system load-frequency control with system parametric uncertainties. The variation bounds of the system parameters are obtained by changing parameters by 30% to 50% simultaneously from their typical values. The controller design procedure is given. The proposed discrete-time robust controller can ensure that the overall mixed system is asymptotically stable for all admissible uncertainties and can achieve satisfactory performance even in the presence of the generation-rate constraint.

discrete-time robust controller can ensure that the overall mixed system is asymptotically stable for all admissible uncertainties. The generation-rate constraint is included in the simulation studies. Comparative studies between the continuous-time robust controller and the proposed robust controller, conducted on an example system, show that the proposed discrete-time robust load-frequency controller can achieve satisfactory performance even in the presence of the generation-rate constraint.
2. P L A N T M O D E L Electrical power systems are complex nonlinear dynamic r;ystems. Since, for load-frequency control, a power system is only exposed to small changes in load during its normal operation, the linearized model can be used and will be sufficient to represent the power system dynamics around the operating point. The block diagram of the linearized plant inodcl is shown in Fig. 2.1.

1. I N T R O D U C T I O N Digital computers are widely being used to implement industrial control systems. In such systems, it is very common to have continuous-time processes and discrete-time controllers. In this paper we are going to design a discrete-time controller for power system load-frequency control (LFC). Discrete-time controller design techniques for load-frequency control problem have been analysed in the last couple of years (12, 14-16]. In most of the existing design procedures for load-frequency control, the uncertainties in the power systems are not taken into account when a controller is designed. The usual design approaches for load-frequency controller employs the linear control theory to develop control laws on the basis of the assumed model. However, an industrial plant, such as a power system, always contains parametric uncertainties. In the design of a controller the uncertainties have to be considered. Otherwise, if the real plant differs from an assumed plant model, a controller designed based on the assumed model may not ensure the stability of the overall system. Recently, [2] and 1171 applied the concept of variable-structure systems (VSS)to the design of load-frequency controllers.

Fig. 2.1 The block diagram of the sample power system

A model we consider here can be written as follows (see [I] and [Z]):

A](t) = -1 A f ( t )t KP A fp(t) -KP A Pd(tX1)

TP TP 1 A&t) = -- A P J t ) - A X p ( t ) TT TT 1 1 AX,(t) = -- A f ( t ) - - A X , ( t ) RTG TC 1 1 -- A E ( t ) t - - ~ ( t ) TG TG TP 1

In this paper, a discrete-time version of a robust controller (see [3]), based on the standard Riccati equation approach, is proposed for power system load-frequency control with system parametric uncertainties. The variation bounds of the system parameters are obtained by changing parameters by 30% to 50% simultaneously from their typical values. The controller design procedure is given. The proposed

(3)

where:

s4f(t) : the incremental frequency deviation in Hz; , 4 P g ( t ) : incremental change in generator output in p.u.

CH3243-3/93/0000-0891$1.00 0 1993 IEEE

MW;

troller

A X g ( t ): incremental change in governor valve position in p.u. MW; A E ( t ) : incremental change in integral control; APd(t) : load disturbance in p.u. MW;

u ( k )= -KDz(k)
such that the continuous-time plant (1)-(4) via the above discrete-time controller is asymptotically stable for all admissible uncertainties. Towards this end, we can employ the discrete-time Iw controller design approaches given in [5-71 to design a discrete-time robust controller. From (5), it can be seen that in the continuous-time case there are only 5 uncertain parameters in the matrices A and B , but when the system (5) is converted into a discrete-time version, all parameters in the matrices F and G will contain uncertainties (in Section 4 the details analysis, conducted on an example, will be given.) so that it is difficult to employ the design approaches given in [5-71 to design a discrete-time robust controller. Here we will propose a simple design procedure to design a discrete-time robust controller for power system load frequency control, which is a discrete-time version of the robust controller proposed in [3]. As presented in [3], the power system ( 5 ) can be rewritten as i ( t )= (A0 AA)z(t) (Bo A B ) u ( t ) (7) where the matrices A0 and Bo are the nominal constant matrices (the nominal parameters can be obtained by choosing the average values in the parameter range.). From the structure of the above model, it is obvious that the matrix AA is of the rank-1 type uncertainty ([3]), i.e.

TG : the governor time constant in seconds;

2'~ : the turbine time constant in seconds; Tp : the plant model time constant in seconds; Kp : the plant gain; R : the speed regulation due to governor action,

Hz

p.u.

MW-l. Here we introduce an integral control of A f ( t )

A E ( t ) = KE /

Af(7)dr

to ensure the regulation property of A f ( t ) ,i.e.

A B ( t )= K E A f ( t )
where K E denotes the integral control gain. The continuous-time model (1)-(4) can rewritten as
X

= Ar(t)

+ Bu(t)+ If A Pd(t)

where
4t) =[ M t ) ,

AP,(t), AXg(t), AE(t)lT

AA=
i=l

A,r,(t)

A, are constant matrices, and

A B = a(t)B
The range of the system parameters are obtained by changing Tp by 50% and TT, Tc and R by 30% simultaneously from their typical values. We then have $ E [ail a111 $f E [Ea 521, $ E [%t a311 E [al a,], E [i%,551. where i i , 3 ( i = 1,...,5) are known constants. and I ri(t) 1 5 f, known.

a(t)

1 5 a, the constants

and a are

Because of the rank-1 uncertainties, Ai can be written as

&

&

Ai = d,eT
where 4 and e, (i = 1, 2, 3) are known constant vectors. Now we introduce the following notations
3

3.

DISCRETE-TIME ROBUST CONTROLLER DESIGN Consider the plant model (5) and set APd(t) = 0. Using the approach given in [4] t o convert the power system model with parametric uncertainties (5) from continuous
time model into discrete time model, we obtain the following discrete-time model

=i

eie;

i=l
The solution of a continuous-time robust load-frequency control design problem for power system (1)-(4) involves solving the following algebraic Riccati equation

z(k

+ 1)

= Fz(k)

Gu(k) = (Fo -6 A F ) z ( k ) (Go

+ AG)u(k)

(6)

PAo+ A i P - P { -( 1- a)BoR;'B? - - T}P+ EIU EQ = 0

1

where the matrices Fo and Go are the nominal contitant matrices, AF and AG indicate the parametric variations in the discrete-time version. The task is to design a discrete-time linear feedback con-

where e, el > 0 and matrices Q and R, are positive definite matrices chosen by the designer.

(8)

Theorem 3.1: ([3]) If for some e, 1 > 0, there exists a positive definite solution P t o (8) then the power system

892

(1)-(4), or equivalently (7), is asymptotically stable for all admissible uncertainties via the following linear feedback law

proposed discrete-time robust load-frequency controller can achieve satisfactory performance even in the presence of the generation-rate constraint (GRC).
4. ANEXAMPLE As illustrated in [2] and [lo], we consider a power system modelled ae (1)-(3) and introduce an integral control modelled as (4).

U(t) =

--R;'B,TP~(~)
(9)

= -Kcz(t)

Proof: See [3] for details.

Since the robust control law (9) can ensure that overall system is asymptotically stable for all admissible uncertainties and the discretization does not change the system stability, we can use discretizing technique to obtain the discrete-time version of the robust load-frequency controller (9) for power system (1)-(4). The discrete-time robust load-frequency controller design procedure is as follow: S t e p 1: Find the range of the parameters for power system (1)-(3) and choose the nominal model by taking the average values of each parameter and decide the bound of the uncertainties; Step 2: Choose the design constants E and el and the design constant matrices Q and &; S t e p 3: Use the algorithm given in [8]to solve Riccati equation (8) and obtain the solution P ; S t e p 4: Construct the feedback gain

As in Section 3, we rewrite the plant model corresponding to (5) i ( t ) = A z ( t ) Bu(t) H A Pd(t)

where A = A0 A A ; E = Bo A B and H = Ho AH. The typical values for the parameters are Tp = 20 sec, K p = 120, TT = 0.3 sec, TG = 0.08 sec, R = 2.4. Then the range of the system parameters can be obtained TT and R by 30% simultaby changing Tp by 50% and TG, neously from their typical values: E [0.033, 0.11, E [4, 121, E [2.564, 4.7621, i!!E [9.615, 17.8571, E [3.081, 10.6391.

&

&

&

1 K c = -R;'BrP
c

A' =

S t e p 5: Find the eigenvalues A, ( i = 1, ..., 4.) for the closed-loop system A0 - BoKc; S t e p 6: Using the approach given in [4] to convert the power system model with uncertainties (5) from continuous time model into discrete time model (the sampling time is T,); S t e p 7: Find the range of the parameters for the discretetime model and choose the nominal model by taking the average value of each parameter; S t e p 8 : Choose the eigenvalues t i for the discrete-timenominal model as t i = exiTa ; S t e p 9: Using Ackermann's Formula ([9]) to find the discretetime feedback gain KD for the discrete-time nominal model. The control laws are as follows:
U(k)

-0.0665 0 -6.86 0.6

8 -3.663 0 0

0 3.663 -13.736 0

0 0 -13.736 0

= -KDz(k)

~ ( t )= u(k)

kT, 5 t < ( l e

+ 1)T.

I ri(t)l i I
It follows that
3

R e m a r k 3.1: Based on [13], we can choose the sampling time T, such that -A,,,T, < 0.5 whxre A ,,

T = Cd,c = dZag(4, 4, 100, 0)

i=l

= maz{A;, i = 1,...,4}.

In next section we will use the design procedure to design a discrete-time robust load-frequency controller for an example system and study the responses of the system for a load change A P d ( t ) = 0.01. Comparative studies between the continuous-time robust controller and the proposed robust controller, conducted on the example system, show that the

U = Ce,eT =
i=l

0.1436 -0.0335 0.1557 0.1557

-0.0335 4.302 -0.302 0

0.1557 0.1557 -0.302 0 0.4718 0.1698 0.1698 0.1698

A B = a(t)Bo and AH = P ( t ) H o . where I a ( t ) 1 5 a = 0.3, I P ( t ) 1 5 P = 0.5.

Choosing
E

= 1, Q = I , el = 5 and R, = 1 and solving he

Riccati equation (8)gives

P=

1.6356 1.4135 0.1427 1.4056

1.4135 4.0188 0.3517 1.0386

0.1427 0.3517 0.1109 0.0878

1.4056 1.0386 0.0878 3.2002

then it follows that

u ( t )= - K c z ( t ) = - A R i ' B T P c ( t )

Remark 4.1: Comparing the plant model in continuoustime case and the discrete-time model obtained above, we can see that in the continuous-time case there are only 5 uncertain parameters in the matrices A and B , but in the l l elements contain matrices P and E, and and nearly a uncertainties so that it is difficult to employ the approaches given in [5-71t o design robust controller. It is much easier to use our design procedure here to find the discrete-time feedback gain.
In the above design and analysis, the effect of generation rate constraint (GRC) is not considered. In real power systems, there exists a maximum limit on the rate of the changein the indicated that generating power. The results in [Ill and [12] the GRC would influence the dynamic responses of power systems significantly. In the case where GRC is considered, the system will present larger overshoots and longer setting times, compared t o the case where GRC is not considered. Furthermore, since the system parameters are unknown, the overall system may become unstable in presence of a load disturbance. In this section we will test the effectiveness of our proposed robust load-frequency controller for a power system with GRC. The GRC of 0.1 p.u. per minute ((11) and [12]) is considered, i.e. the GRC is

where

1 Kc = -R;'B,TP = [IC',

h, Ic3,

I C 4 1

= I1.893, 4.762, 1.516, 1.6581

The eigenvalues for the closed-loop system A0 as follows: eig(Ao - BO&)

- BoKc are

= lh, Xz, X3, A41 = [ -25.147, -9.339, -2.840, -0.9621

According to Remark 3.1, we have T, < 0.0199 aec. Using the approach given in [4], we can find the discrete-time models as follows (the sampling rate is T, = 0.01 sec):

F=

0.99967 0.11846 0.00272 -0.00037 0.97468 0.04432 -0.02871 -0.00058 0.90832 0.00600 0.00036 0.00001

-0.00002 -0.00118 -0.09167 1.00000

G = [0.00016, 0.00395, 0.16353, O.OOOOOf*

and

F= -

0.99890 0.03904 0.00048 -0.00242 0.95340 0.02317 -0.10170 -0.00614 0.83637 0.00600 0.00012 0.00000

-0.00016 -0.00395 -0.16353 1.00000

O.OOOOO]'

G = (0.00002, 0.00118, 0.09167,

1 1

nPg5 61= 0.0017 p.u. MW/sec

The GRC is taken into account by adding a limiter to the turbine and also adding to the integral control part to prevent the excessive control action. The block diagram of the plant model with generation rate constraint (GRC) is shown in Fig. 4.1.

The matrices F and E, and G and G. give the upper bounds and low bounds of the parameters in discrete-time cases. and E, Choosing the average values of each element in and G and gives the discrete-time nominal model of the plant, Fo and Go,

F0=

Fig. 4.1 The block diagram of the sample power system with GRC The schematic diagram of the overall mixed system is given in Fig. 4.2

0.99928 0.07875 0.00160 -0.00009 -0.00140 0.96404 0.03375 -0.00257 -0.06520 -0.00336 0.87234a -0.12760 0.00600 0.00024 0.00000 1.00000
for nominal discrete-time

G o = [0.00009, 0.00257, 0.12760, O.OOOOOf*

The desired eigenvalues z, = model are found to be
[zI
22

z 3

241

= [0.777 0.911

0.972

0.9901

Using Ackermann's Formula ([9]),the discrete-time feedback gain KO can be found as

KO = [1.674 4.340

1.360

1.4211

894

Fig. 4.2 The schematic diagram of the overall mixed system

Time (sec)

Consider the case where the load change is 0.01 pu., i.e. APd = 0.01. Case 1: We choose three different groups of system parameters in the parameter region. Curve I: with nominal parameters; Curve 2 : = 0.1 ! k TT 1 - = 4.762, $ = 17.857, T P = 12 lG = 47 1 7 Tp TT = 2.564' 2 TO = 9.615, 1 = 3.081. RTc The system responses, Af(t), for the proposed discretetime robust controller are shown in Fig. 4.3, where GRC SI == 0.0017 and 62 = 0.00005.

A f ( t )responses for different controllers

& & = 10.639; Curve 3: & = 0.033

i
Time (TOC)

Tim (IOC)

Fig. 4.5 Af(t) responses for different cohtrollers

Remark 4.2: F'rom the results in Case 1 we can see that

the proposed discrete-time robust controller is effective and it can ensure that the overall system is asymptotically stable for all admissible uncertainties. The simulation study of a numerical example indicated that for our example system the proposed robust controller can achieve good performance even in the presence of generation rate constraint.

Fig. 4.3 Af(t) responses for different parameter variations Case 2: Next we are going to compare the proposed discretetime controller with the robust controller proposed in [3]. Curve 1: The proposed discrete-time robust controller; Curve 2: The continuous-time robust controller proposed in

Remark 4.3: From the simulation results in Fig. 4.4 and Fig. 4.5, we can see that the incremental frequency deviation A f ( t ) responses experience larger overshoots and
longer setting times when the system is controlled by the discrete-time robust controller than by the continuous-time r'obust controller given in [3].
5. CONCLUSION I n this paper, a new discrete-time version of the robust loadis proposed to improve the frequency controller given in [3] power system performance. The variation bounds of the system parameters are obtained by changing parameters by 30% to 50% simultaneously from their typical values. The controller design procedure is given. Simulation studies show that the advantages of the proposed controller are as follows: Firstly, the proposed discrete-time robust controller i s a

[31,
Fig. 4.4 shows the responses of A f ( t )for two different controllers with the plant parameters: = 0.1, = 12, = 10.639 and Fig. 4.5 = 4.762' Tc = 17.857, shows the responses of A f ( t ) for two different controllers = 0.033, = 4, = 2.564, with the plant parameters: - = 9.615, = 3.081.

&

&

&

TG

&

Y95

discrete-time version of the continuous-time robust controller presented in [3]. The design procedure given in this paper is much easier to use to design a discrete-time robust controller for a continuous-time plant with parametric uncertainties. When we design the controller, the true system parameters are not needed, the knowledge required for the plant is only the bounds of the plant unknown parameters. Secondly, the proposed robust controller is effective and it can ensure that the overall mixed system is asymptotically stable for all admissible uncertainties. Our simulation results show that for the example system the proposed discrc te-time robust load-frequency controller can achieve good performance even in the presence of generation rate constraint. Finally, the controller design procedure proposed in this paper can be employed to design a computer-based control system for power system load-frequency control.

for power system load-frequency control, IEEE Trans. Power Systems, PWRS-4, pp.122-128,1989.
[ll] W. C. Chan and Y. Y. HSU,Automatic generationcontrol of interconnected power systems using variablestructure controller, Proc. IEE Pt. C, Gen., n u n s . and Distrib., Vol. 128, pp.269-279, 1981. [12]T. Hiyama, Optimasation of discrete-type load-frequency regulators considering generation rate constraints, Proc. IEE Pt. C, Gen., %ns. and Distrib., Vol. 129, pp.285-289, 1982.

[13] N. K. Sinha and Q. J. Zhou, Discrete-time approximation of multivariable continuous-time systems, Proc. IEE Pt. D. Control Theory and Appli., Vol. 103, No. 3, 1983. [14] M. L. Kothari, P. S. Satasngi and J. Nanda, Sampleddata automatic generation control of interconnected reheat thermal systems considering generation rate constraints, IEEE Duns. Power App. and Syst., PAS100, pp.2334-2342, 1981. [15] J. Nanda, M. L. Kothari and P. S. Satasngi, Automatic generation control of an interconnected hydrothermal system in continuous and discrete modes considering generation rate constraints, Proc. IEE Pt. D. Control Theory and Appli., Vol. 103, pp.17-27,1983. [16] A. Kumar and 0. P. Malik, Discrete analysis of loadfrequency control problem, Proc. IEE Pt. C, Gen. %ans. and Distrib., Vol. 131, No. 4, pp.144-145, 1984. [17] A. Kumar, 0. P. Malik and G. S. Hope, Discrete variable structure controller for load frequency control of multiarea interconnected power systems, Proc. IEE Pt. C, Gen. Trans. and Distrib., Vol. 134, pp.116122,1987.

REFERENCES:
[l] R. N. Dhar, Computer aided power system opemtion

and analysis, Tats McGraw-Hill Publishing Company Ltd, New Delhi, 1982.
[2] A. Y. Sivaramakrishnan, M. V. Hariharan and M. C. Srisailam, Design of variable-structure load-frequency controller using pole assignment technique, Int. J. Control, Vol. 40, pp.487-498,1984. (31 Y. Wang, R. Zhou and C. Wen, Robust controller design for power system load frequency control, Proc. 1st IEEE Conf. Control Applic., pp.642-646, Dayton, Ohio, USA, 13-16 September, 1992. [4] W. K. Chua and S. M. Fong, Computer-aided loadfrequency controller design, Report, School of Electrical and Electronic Engineering, Nanyang Technological University, 1993. [5] K. Furuta and S. Phoojaruenchanachai, An algebraic approach to discrete-time H" control problems, Proc. 1990 American Control Conference, pp.2067-2072, San Diego, CA, May 1990. [6] D. J. N. Limebeer, M. Green and D. Walker, Discrete time H" control, Proc. 28th IEEE Conf. Decision and Control, pp.392-396, Tampa, FL, December 1989. [7] M. Fu, C. E. de Souza and L. Xie, Quadratic stabiliz& tion and H" control of discrete-time uncertain systems, Proc. Int. Symposium Math. Theory of N e t works and Systems, Japan, 1991. [8] I. R. Peterson, A stabiliiation algorithm for a class of uncertain linear systems, Syst. Contr. Lett., Vol. 8, pp.351-357,1987. [9] K. Ogata, Discrete-time control systems, Prentice-Hall, 1987. [IO] C. T. Pan and C. M. Liaw, An adaptive controller

896