You are on page 1of 6

EDF EXPERIENCE WITH CO-ORDINATED AVR+PSS TUNING

T. Margotin*, H. Bourl&s**,Senior Member, IEEE


* Electricitt de France, Division Recherche & Dtveloppement, 1 avenue du GCnCral de Gaulle, 92141 Clamart, France ** Scientific consultant for EDF and Professor at the Conservatoire National des Arts et Metiers, Paris, France
Abstract - State feedback voltage controllers including both active power and rotor speed stabilizing loops have been successfully used for many years on the French power system: the "Four Loop Regulator" (FLR) is currently installed on most of the French biggest power plants. It will be replaced in the next years by a robust coordinated AVR+PSS, called the "Desensitized Four Loop regulator" (DFLR), having a good performance in terms of transient and small signal stability. It is shown here that the DFLR can be tuned with few design parameters, using a systematic procedure. The opportunity to transform the specific structure of this controller into standard IEEE AVR+PSS structures has been proved. A two step method is therefore proposed for tuning the parameters of a standard IEEE AVR+PSS: (i) design a DFLR, (U) make the above-mentioned conversion, which depends on the type of the excitation system. The AVR+PSS obtained in this manner are efficient for damping all kinds of electromechanical oscillations of a large scale power system (including inter-area oscillations). This approach should be combined with modal analysis techniques to determine the best location of the AVR+PSS to be installed or retuned. Its efficiency has been verified on several large scale systems and is illustrated here on a modified version of the New England system. The obtained AVR+PSSs have a good robustness against significant changes in the characteristics of the network.

rotor speed generates, due to its positive feedback, an overexcitation during and immediatelv after a three phase fault. DFLR will be implemented in place of FLR in the next years. More recently, it was shown that DFLR is a concept to be exported outside France. Although its structure is not classic, as it is a state feedback control, it can be put in several classic AVR+PSS IEEE DCIA+PSSIA [IO]and structures: IEEE STlA+PSSIA [4], other ones. In this manner, the structure of the obtained controller is classic (and, therefore, it can be easily implemented), but so is not the tuning of the parameters. It has indeed several features: - Due to the design method, the AVR and the PSS have a coordinated action. - The PSS has a large gain, a fact which is also related to the coordination. This makes it possible to stabilize the power system in very different situations, due to the robustness of the obtained controller. - A price to pay for this robustness could be a poor behavior in case of, e.g., islanding. The usual limitation of the PSS can indeed be insufficient. This is why at EDF DFLR is complemented by protections like disconnection of the PSS in case of islanding, or terminal voltage limiter. With these cautions, the large gain of the PSS is definitively an advantage, as far as the power system stability is concerned. - As explained in the sequel, due to the fact that DFLR has a good performance in case of a generator feeding a long line, it is also efficient for damping low frequency inter-area oscillations [4, IO],although its main function is to damp out local ones. The paper is organized as follows: the DFLR is presented in Section 11, where the main ideas of [SI are recalled: the design method is described and an example applying the method is presented. In Section 111, it is shown how the DFLR can be converted to standard AVR+PSS structures and the performance of the resulting controller in terms of small signal stability is illustrated by simulation results in the case of the modified New England test system affected by an inter-area mode. Section IV is devoted to concluding remarks.

Keywords - AVR, PSS, Linear quadratic control, Desensitivity, Inter-area oscillations

I. INTRODUCTION EDF has a long experience in designing and tuning automatic voltage regulators (AVR) and power system stabilizers (PSS). The Four Loop Regulator (FLR) has been implemented on almost all French nuclear power plants since the beginning of the 80's. This controller was designed using a state space approach (pole placement) [6].Compared with classical AVR+PSS, this controller allows to stabilize a power plant that feeds very long lines. Such a performance (concerning both small-signal and transient stability) is a need for the French power system where several power plants are located far from consumption centers [2]. As was shown in the ~ O ' S this , controller is not optimal in case of large variations of the network frequency. It induces indeed a large coupling between frequency and terminal voltage. In order to overcome this problem, the Desensitized Four Loop Regulator (DFLR) was then designed [SI. The design method, also based on the state space representation, uses optimal control and robustness theory. DFLR enjoys the same performance as FLR when feeding long lines, but its behavior is globally better, as the abovementioned coupling between frequency and terminal voltage has been much decreased and made very reasonable. As the FLR, it provides good transient stability performance, as the increase of the

11. A ROBUST COORDINATED AVR+PSS: THE DFLR

The aim of this section is to describe the design method of DFLR. It is based on the single machine-infinite bus model in state space form. This model is augmented for obtaining a control law with integral term. The desensitivity method is then applied. We explain first this theory, then how the designer should adjust the design parameters based on the table of robustness margins, and finally an example of design is presented.

0-7803-5569-5/991$10.00 0 1999 IEEE

47

A. Basic linear model The model which is considered for the design is the classical system "single machine, infinite bus". This model is linearized around an operating point defined by specific values P, V, q' and X of the active power, the terminal voltage, the reactive power and the external reactance, respectively. That model is reduced to order 3 using the balanced realization technique, which can be viewed as a principal component analysis method for dynamic systems [9]. For any signal 5, let 6( denote the deviation of ( from its steady-state value (when it exists). The state is chosen as being z = [SV 6P where o is the rotor speed; the control U is &E,. Note that for the design of DFLR, the exciter is supposed to be compensated (on the French power system AC type exciters are compensated by a lead network). The variable to be controlled can be written 6V = C z, with C = [ l 0 01. Set 0 = [P V q XIT; the matrices of the linearized model depend on 0, i.e., the differential equation of the linearized system around the operating point 8 is of the form

The block-diagram in Fig. 1 clearly shows that the DFLR can be viewed as a coordinated AVR+PSS; U, can be viewed as an additional stabilizing signal. The gains K,, Kn, K, and K, are constant.

C. Desensitivity Desensitivity is a method for designing a robust controller with respect to parametric uncertainties, i.e., a controller stabilizing system (3) with a suitable level of performance, not only when 0 has its nominal value O0, but for every 0 in 1 ) . In order to clarify the main idea of desensitivity, it is assumed in this section that 0 is a scalar parameter (this is generally not true for the AVR+PSS design problem). 1 ) The state x is a function o f t and 0, and so is also the control U by ( 4 ) , i.e. x = x(t, e),U = u(t, 0 ) Let O0 be a nominal value of 0. Desensitivity consists in minimizing a quadratic index involving x$t) = x(t, 00), U0(t) = u(t, 02, and also the partial derivatives to(t) = &x(t, 8 ) and po(t) = deu(t, 02 (where de denotes the partial derivative with respect to 0);c$t) and p$t) are called the "sensitivities". This index is defined by

z = F(0) z + G(0) U (1) It is assumed that the vector of parameters 0 is constant with respect to time and unknown (thus, it is a considered as random vector) but belongs to a known domain 9, called the "admissible parametric domain".
B. Augmented model and control structure Let V, be the terminal voltage reference (which is assumed to be constant). In order to satisfy the steady-state objective, the plant model (1) is augmented with the additional state e defined by e=V-V, (2) With the control synthesized below, e(t) tends to a constant as t tends to infinity, thus the regulation error V(t) - V, tends to zero; as a result, the steady-state value of V(t) is Vc, so that 6V = V - Vc. In addition, as e is defined up to a constant, one can set 6e = e. The "augmented state" is x = [zT elT. Equations (I), (2) yield x =A(@) x + B(0) U where (A(@), B(0)) is controllable and is defined by A(@)=[ F(0)

(3)

where d is the variance of 0 (in case when 0 is a vector, it should be replaced by a covariance matrix [5]). The minimization of sensitivities ensures that the "perturbed The control law is constrained to be a classic state feedback trajectories" x(t, 0 ) and u(t, 0) remain close to the "nominal (4) trajectories" x$t) and u,(t) when the current parameter value 0 is u=-Kx and therefore its scheme is the one presented in Fig. 1. slightly different (and, in practice, rather different) from 0,. In this manner, a good robustness of the performance is obtained. The Notice that the active power P has been replaced by the larger is (3, the more weighted are the sensitivities, and the more "accelerating power" P - Pm, where P, denotes the mechanical improved is the robustness. The price to pay for this improvement, power supplied by the turbine (in practice it is estimated from the however, is a deterioration of the performance, so that a good first stage pressure). compromise should be found.

] B(O)=[

GT' ]

This notation is not classic, but it allows one to avoid a confusion with the state weighting matrix (see ( 5 ) ) .

'

48

2) Differentiating (4) with respect to 8 yields F&t) = - K {Jt). Now, differentiating (3) with respect to 8 and using the latter equation and as well as (3) one obtains

x, =

e, [ x i e,.]

(7)

where the expressions of A,,B,,C, can be easily determined; the two former matrices depend on K. Thus, assuming that K is known, a controller can be synthesized for system (6), (7) using the LQG theory [I]. However, K is the controller gain matrix we want to calculate, hence iterations should be used. At step 0, a first gain matrix KO is calculated, such that (4) stabilizes the nominal system (3) (with 0 = 02. For this, the classical LQ method [ l ] is used: the quadratic index (5) is minimized with CJ = 0; the matrix Q is chosen diagonal; KOis the gain of the non-desensitized LO regulator. At step 1, sensitivities are for the first time taken into account by choosing CJ >O. The resulting quadratic index is minimized for system (6), (7) with K replaced by KO, using LQG theory; the order of the LQG controller then obtained in nonzero, hence it is reduced to order zero using, e.g., the method developed in [7]*. In this manner, a first desensitized controller with gain matrix K, is obtained. Step 1 can now be repeated with KOreplaced by K,, etc. In this manner a sequence of gain matrices (K) is determined. The procedure ends when KO= Kn+, to within the tolerance. The number of iterations does not generally exceed about 10.

For a correct calculation of the open loop margins, the loop should be open at the suitable place, which is the input of the control system, indicated by a cross in the scheme in Fig. 2. The model errors can indeed be assumed to be gathered at this place. Note that the resulting open loop system is then a mono-input-mono-output one, hence having a scalar transfer function, whose usual Bode and Nyquist plots can be considered. The delay margin is the ratio of the phase margin (expressed in rad.) by the cutoff frequency (expressed in rads) assuming that the latter is unique. It is the maximum delay which can be inserted at the place indicated by the cross in the closed loop system without destabilizing the latter. This margin characterizes robustness against sampling delay and unmodelled dynamics. The modulus margin is the distance between the Nyquist plot of the open loop transfer function and the critical point -1. A sufficient modulus margin ensures a good robustness against uncertainties on gains and against neglected non-linearities. The closed loop system generally has several pairs of complex poles, each of them having a damping ratio. In the tables presented in the sequel, the damping of the closed loop system is defined as the worst of these ratios. Generally speaking, this damping is very significant as far as small signal stability is concerned. The table gathering these margins at various operating conditions is a key tool for the design, as shown by the following example.

E. Example
1) Parametric domain and design Doint In what follows, all quantities are expressed in p.u. except for w which is in rad/s; X includes the reactance of the transformer. Let the admissible parametric domain 1)be: P = 1 (Pn base), 0.95 I V 5 1.05 (Vm base), -1 I q I 1 (4. base), 0.2 5 X 5 1 (Snbase) In the tables below, the variations of q are not presented due to the lack of place. Notice that 1)is very large and that V and X are the key parameters. The design point is chosen as 00:= [POVO qo &IT = [l 0.95 1 O.6lT; Vo has been chosen at its lowest value , is the mean of X. because it is the most constraining situation, X Recall that the state is x = [SV 6P 6w elT and that the input is U = 6E,. 2) Non desensitized LO controller desim ( CJ = 0)

D. Table of robustness margins and their use for adjusting the design parameters
Of paramount importance for the design are the robustness margins at representative operating conditions. Two types of margins are considered: - Those calculated from the open loop: the delay margin and modulus margin [3]. The damping of the closed loop system.

Generator

+ Exciter

Infinite bus

Stabilizing signals

F J ~ z / J

Where the loop should be opened

The state weighting matrix is chosen as Q = diag (4000, 0, 0, 1000). The idea leading to this choice is the following: only the states 6V -the variable to be controlled- and e -the output of the integrator- are weighted, in order to obtain a good PI controller for the voltage regulation (i.e., a good proportional and integral controller). The controller gain matrix is

K,,= [ 74.7 35.4 -5.28 31.6 ]


According to the authorsexperience, that reduction to order zero does not significantly disturb the closed loop dynamics, as far as AVR+PSS design is concerned. However, in [5], that reduction to order zero is made at the final step only. The latter method is more accurate. The reduction to order zero is assumed to be made at each step here for the sake of simplicity.

The robustness of this controller is not sufficient, because the system is not stabilized in the whole admissible parametric domain 1 ) . as shown in Table 1.

49

Notice that a poorly damped closed loop system is obtained for large values of X. Even if the "real" equivalent reactance connecting the generator to the rest Of the power system (calculated from the short circuit Power) is not large, the behavior of the as it concerns controller should be improved for large values of the low frequency domain. Therefore, this controller should be desensitized with respect to this parameter.

3) Desensitized controller desirrn


The standard deviation (J is increased until satisfying damping is obtained for all values ofx X. Choosing ax=O.Syields the results given in Table 2. considered operating Good stability margins are obtained at conditions. Therefore it is not necessary to desensitize the controller with respect to critical parameter V in this particular case. Note that the described procedure is very systematic. The following controller gain matrix is obtained: A K = [ 76.2 64.3 -10.1 29.81 = [ Kv K, K, K,] (8) Basically, the gains of both stabilizing signals P and o have been increased.

x,

1 1 1 . CONVERSION OF THE DFLR TO A STANDARD

A m + Pss STRUCTURE
A. Analysis of the structure of DFLR From Fig 1, the control law of the DFLR can be written where
U , ,

(9) U =U , , +U , is generated by a PI controller with transfer function K AVRi(s) = K , +e (10)


S

and where U,,& is an additional stabilizing input


U&d= - K m W + KpPa

Table. 1 Robustness margins with the non desensitized controller

Assuming that the mechanical power slowly varies, this signal can be seen as the output of a PI controller feeded by the active power, due to the equation
W(t)=

-11

2H

Pa ( t )d t

where Pa = P, - P denotes the accelerating power. The transfer function of this controller is
2Hs Therefore, DFLR can be converted to standard P-input .AVR+PSS structures.

K, PSS,(S) = K, --

B. Example: IEEE DClA

+ PSSlA

In case of a static excitation system, it has been detailed in [4] how the DFLR can be converted, using suitable approximations, to the IEEE STlA+PSSIA structure. It is shown in [lo] that the idea used in [3] is quite general and can be applied in other cases. To illustrate this, a DC excitation system is considered, and more specifically the DClA AVR below. The calculations are detailed in [lo]. The structure of IEEE DClA+PSSIA is shown in Fig. 3.

Table. 2 Robustness margins with the desensitized controller

50

AVR(s)

The Bode diagrams of AVR(s) and AVR,(s), defined by (lo), have the shape shown in Fig. 4. Therefore, AVR(s) can be viewed as the approximation of AVR,(s) on the suitable frequency range I 1. A similar rationale can be made for the PSS on the same Ts

[%*

frequency range.

Fig. 3c: Structure of IEEE PSSlA+DClA

Iv. APPLICATION EXAMPLE: THE MODIFIED NEW ENGLAND SYSTEM


A. System description
The New England test system is classic (see, e.g., [ l l ] ) and is shown in Fig. 5 (it is a constant impedance load model, and loads are connected to underlined buses). A 0.26 Hz inter-area oscillation has to be damped. From the small signal stability analysis, a PSS should be installed on machine 9. For this machine, a DFLR has been designed and then converted to DClA+PSSlA structure. The performance of the obtained controller is illustrated by the simulation below. It has been made using the time simulation software EUROSTAG [SI. A 50 ms three phase fault has been made on line 3-18, which is then disconnected. The voltage at bus 18 and the active power in the inter-area link are shown in Fig. 6. Note that the power system is well damped with the PSS, although the topology of the network has changed. This is also true when significant changes are made in other parameters of the power system (such as load characteristics). This proves that the PSSs have a good robustness.

--+

Fig. 3b: Block diagram of AVR(s1 Fig. 3c: Block diagram of PSS(s1

........

Inter-areamode

Inter-machinemode

i.---..;

gain (dB)

Fig. 4: AsymDtotic Bode Diagrams of AVR and AVR,

Fig. 5: New England test system

51

1.14

PU

,.-

;...

...............

.................

........................

REFERENCES
[l] B. D. 0. Anderson, J. B. Moore, Optimal Control: Linear Quadratic Methods. Englewood Cliffs, NJ: Prentice-Hall, 1989. [2] R. BCnCjean, P. Blanchet, J. P. Meyer, P. Hugoud, "A New Voltage Regulator for Large French Alternators", CIGRE Conf., vol. 32-13, 1982. [3] H. Bourlb, F. Aioun, "Approche H, et p-synthkse", in La robustesse -Analyse et synthPse de commandes robustes, A. Oustaloup, Edt., Paris: Hermks, pp. 163-235, 1994. [4] H. Bourlks, S. Peres, T. Margotin, M.P. Houry, "Analysis and Design of a Robust Coordinated AVRPSS", IEEE Trans. on Power Systems, vol. 13, n"2, pp. 568-575, 1998. [ 5 ] A. Heniche, H. BourlBs, M. P. Houry, "A Desensitized Controller for Voltage Regulation of Power Systems", IEEE Trans. on Power Systems, vol. PWRS-10, pp. 1461-1466, 1995. 161 E. Irving, J.P. Barret, C. Charcossey, J.P. Monville, "Improving Power Network Stability and Unit Stress with Adaptive Generator Control", Automatica, vol. 15, pp. 31-46, 1979. [7] D. Kavranoglu, "Zeroth Order H, Norm Approximation of Multivariable Systems", Numer. Funct. Anal. and Optimiz., vol. 14, pp. 89-101, 1993. [8] B. Meyer, M. Stubbe, "EUROSTAG, a Single Tool for Power System Simulation", Transmission & Distribution International, March 1992. [9] B. C. Moore, "Principal Component Analysis in Linear System: Controllability, Observability, and Model Reduction", IEEE Trans. Automat. Contr., vol. AC-26, pp. 17-31, 1981. [lo] H. Quinot, H. Bourlb, T. Margotin, "Robust Coordinated AVR+PSS for Damping Large Scale Power Systems", IEEE PE455-PWRS-0-12- 1998. [ l l ] D. Wong, G. Rogers, B. Porreta, P. Kundur, "Eigenvalue Analysis of Very Large Power Systems", IEEE Trans. on Power Systems, vol. PWRS-3, pp. 472-480, 1988. BIOGRAPHIES

Voltage at bus 18

Time (s.)

450.

'

16.

'

26.

30.

46.

Active power in the inter-area link

51. Time(s.)

Fig. 6: 50 ms three ohase fault on line 3-18 IV. CONCLUDING REMARKS

The Desensitized Four Loop Regulator which design method is presented in section I1 of this paper, has a good performance in Thibault Margotin received his Engineering Degree from Ecole terms of transient and small signal stability. As shown in section Centrale de Nantes (France) in 1994. He is a Research Engineer at 111, approximations allow to convert the particular structure of the the EDF Research and Development Division located in Clamart, DFLR to standard IEEE structures as the AClA+PSSlA, France. e-mail: thibault.margotin@edf.fr. DClA+PSSlA, STlA+PSSlA. This conversion lead to a similar behavior of the converted controller and of the DFLR on a suitable Henri Bourlss received his Engineering Degree from Ecole Centrale frequency range. The controller resulting from the conversion has de Paris (France) in 1977 and his Ph. D. from the Institut National the same ability as the DFLR to improve the damping of an inter- Polytechnique de Grenoble (France) in 1982. He worked at the EDF area mode, as far as the concerned machine has a sufficient Research and Development Division as Senior Research Engineer participation to the mode. A reason for this is that the DFLR is until 1997. He is currently a scientific consultant for EDF. On the designed to be able to stabilize the closed loop system, even in the other hand, he is Professor of the Chair of Automatic Control at the case of a large value of the external reactance connecting the single Conservatoire National des Arts et MCtiers where he is also the machine to the infinite bus in the design model (the frequency of the Director of the Laboratoire dAutomatique des Arts et MCtiers, Paris, electromechanical mode of the machine is then low). T h s property, France. e-mail: Henri.Bourles@paris.ensam.fr. obtained by desensitizing the controller with respect to X, leads to a stabilizing behavior in case of inter-area mode, as confirmed by the simulations presented in section 111.

52

You might also like