Professional Documents
Culture Documents
453
i = 1 , 2 , . . . , n,
1
step to establish the weighting matrix optimization model is to
obtain the sensitivity values of the closed-loop system Since S': has been obtained from eqn.(6), the sensitivity
eigenvalues A, the feedback gain K and the eigenvalues p of the matrix S4", of the feedback gain matrix K to qij can be calculated
weighting matrix Q to the elements qlJ of the weighting matrix directly from eqn.(3):
Q. They are denoted by S& S E and Sg respectively.
S4",= - R-I BT SqUp (12)
Suppose S': is the sensitivity matrix of the solution matrix
P of the Riccati equation to the elements qlJ of the weighting The sensitivity matrix St can be re-arranged in the
matrix Q. From eqn.(4) it can be derived that S': satisfies the following form:
following Liapunov
aqij where
i = 1,2;. . , n,;
1
i Ij 5 n, ak.. ak.. . . . L
ak.. L
ak.. ...L
ak- . . . - akjj
22
If A F is used to represent the system matrix of the closed- 3411 3412 aqln, 3422 aqb, aqn,n.,
loop system
i = 1.2;. . , n,; j = 1,2;. . , n ,
AF = A - B R-' B ~ P (7)
according to Hellman-Feynman Theorem, the sensitivity value Supposing j3 is the eigenvalue vector of the weighting
akk/apjj of the eigenvalue & of AF to pij is given in the matrix Q, from Hellman-Feynman Theorem the sensitivity value
following form [lo]: apk/aqij of the eigenvalue p, to qij is
where 4 and 2, are the eigenvectors of the eigenvalue kk where f?k and Vk are the eigenvectors of the eigenvalue p k
corresponding to A: and A, respectively. For every eigenvalue Corresponding to QT and Q respectively. Similar to eqn.(lO), the
a k ( k = 1, 2 , . . . , n , 1, from eqn.(8) the sensitivity matrix S)
sensitivity matrix SQ of j3 to Q has the following form:
can be formed as: r 1T
a& -ali,
- . . . - where
apIl aP12 a~ I n ,
ah, -a&
- . . . -
ap21 JP22 aPh,
. . . . . (9)
i = 1.2, . . . , n ,
--...-
. ... I . ..
apnfil.
1I
3. OPTIMIZATION MODEL AND PROCEDURE
Real parts of the dominant eigenvalues of the closed-loop
system determine the decreasing rate of the dynamic oscillations,
Once the sensitivity matrices S$ and S,: are obtained, the and the imaginary parts determine the frequency of the dynamic
sensitivity value a)ik/aqij of the eigenvalue lit to qij can be oscillations. In general, the dynamic performance of the control
calculated by using the derivative principle of compound system mainly depends on the real parts of the dominant
455
eigenvalues. Thus in this paper, the optimization objective For linear optimal control systems, the weighting matrix has
function and the constraint consideration are based on the real its specific significance. It is used to constrain some state
parts. However, the optimization model can be easily modified to variables. Thus it may be necessary to demand that some
consider the effects of the imaginary parts. elements qr, of the weighting matrix have some specific values.
Mathematically, the objective of introducing an optimal Simultaneously, in order to enhance the flexibility of the
control system is to improve the eigenvalue locations of the proposed method to be able to optimize the weighting matrix not
closed-loop system in the complex plane["'. The objective of only in the full- matrix form but also in the diagonal form, the
optimizing the weighting matrix Q is to move the dominant following constraints are needed :
eigenvalues of the closed-loop system towards the left side in the
@ I Q + AQ I @ (20)
complex plane as far as possible without violating the given
constraints. Without loss of generality, suppose ,.A is the real part As a matter of fact, these constraints are not absolutely necessary
of the dominant eigenvalues of the closed-loop system. The for the weighting matrix optimization in the full-matrix form. It
objective function of the optimization model can be chosen as provides an option for the designer to decide whether to use
them or not, and makes this method more flexible.
Min M l = S $ A Q (16) The relation between the eigenvalues A of the closed-loop
AQ
system and the weighting matrix Q is nonlinear. But for
where MI is the amount of movement of the dominant simplicity, a linear model is used in the proposed optimization
eigenvalue with respect to the change AQ of the matrix Q. technique. To guarantee that the linear optimization model is
For the sub-dominant eigenvalues, they may be moving applicable to solve the nonlinear problem, the optimizing step
towards the right side in the complex plane while the dominant AQ must be constrained so that in each optimization iteration the
eigenvalues are moving towards the left in the optimization proposed method works within a linear region.
procedure. In order to avoid this undesired phenomenon, the
following constraints are added in the optimization procedure AQ 5 L\Qo (21)
mrl -
456
0.081 0
+ 0.051
0
0
0
0
0
[AvjAE,
(24)
handled by the proposed method in the same manner. If a point the dominant closed-loop eigenvalue may increase in the
solution is obtained, the proposed method will give the best process of making the weighting matrix definite as seen in
closed-loop eigenvalue locations which satisfy the constraints. Fig.2(a). The damping ratio of the closed-loop system also
Otherwise, a state observer must be used to get the estimates for decreases corresponding to the dominant closed-loop system
the missing state variable(s). eigenvalues as shown in Fig.2(b). These results show that with
Closed-loop dynamic response to an impulse disturbance on the proposed optimization technique, a proper weighting matrix
the state variable m at 1 s with different controllers is shown in can be easily obtained to meet the demand of the desired
Fig.1. This figure clearly shows the effectiveness of the proposed locations of the closed-loop system eigenvalues in the complex
weighting matrix optimization technique. plane.
_.
.. U
a i 2 4 s 6 (a) Convergence of the closed-loop eigenvalues
- with reduced-ordersystem designed by the proposed method
---- with full-statefeedback control system
-- with reduced-ordersystem designed by the MEE method
Fig.1 Impulse responses with different controllers
e 1 2 3 4 5 6
E
459
York, 1977
- with
- -- - with
proposed PSS
conventional PSS [ 131 Murtagh, B.A., 'Advanced linear programming', McCraw-
-- without stabilizer Hill International Book Co., New York, 1981
8. APPENDIX
8.1 System model
6=co&l
Mb=T,,, -KdO-T,
T ; o . k i = E ~ f- ( x d - x & ) i d -Ed
TA = KA ( V,q - V, + U ) - Er
Time s
- : : - : - - - : : : : : . : : : - . + . . - - : : : : . k 0 = v d + r, id - xq i,
a 1 2
Fig.4 Step response with leading power factor operation Eq = Vq + ra iq + xd id
T, = E d i q - ( x i - x q ) i d i q
6. CONCLUSIONS :v = vd' + v;
With the proposed weighting matrix optimization technique, vd = V, sin6 + re id - x, i,
one can easily get a suitable weighting matrix to ensure that the V, = V, cos 6 + re iq + x, id
desired closed-loop eigenvalues to be achieved. Also, this
optimization technique can consider many kinds of constraint. 8.2 Conventional PSS model
The studies described in this paper show that a reduced-order
feedback controller can be designed with the elements of the
weighting matrix chosen in an optimal and coordinated manner. 8.3 System and CPSS parameters
This gives the weighting matrix a new significance.
r, = 0.0031, xd = 1.6, X& = 0.17, X, = 1.53
7. REFERENCES Tdo = 5.18, M = 5.0, Kd = 0.0
Gibbard, M.J., "Co-ordinated design of multimachine power re = 0.05, x, = 0.5, K, = 50.0, T, = 0.01
system stabilizers based on damping torque concepts", IEE Kg=0.8, T ~ = 2 . 5 , T,=0.15, T2=0.05
proceedings, V01.135, ptC, No.4, 1988, pp.276-284 All resistances and reactances are in pu and time-constants
Raghavendra, K.R. and L. Jenkins, "Design of a near in sec.
optimum regulator for a power plant", Electric Machines 9. BIOGRAPHY
and Power Systems, Vol.10, 1985, pp.1-13
Chen, Gan-Ping (S'91) was born in Sichuan, P.R.China, on
Chan, W.C. and Y.Y. Hsu, "Optimal control of electric January 12. 1964. He received his B.Sc. degree from Chengdu
power generation using variable structure controllers", University of Science and Technology, Sichuan, P.R.China in
Electric Power Systems Research, V01.6, 1983, pp.269-278 1984 and his M.Sc. degree from Chongqing University, Sichuan,
Cheng, S.J., O.P. Malik and G.S. Hope, "Damping of P.R.China in 1987. both in electrical engineering. Presently, he is
multi-modal oscillations in power systems using a dual-rate working towards his Ph.D degree in the Dept. of Electrical and
adaptive stabilizers", IEEE Trans. on Power System, V01.3, Computer Engineering, the University of Calgary, Canada. His
No.1, 1988, pp.101-108 fields of interest are power system stability control with optimal
IEEE Committee Report, "Dynamic models for steam and and adaptive excitation controllers, and their real-time
hydro turbines in power system studies", IEEE Trans. on implementation; power system operation, analysis, and planning.
PAS, Vol.PAS-92, No.6, 1973, pp.1904-1915
Malik, O.P. (M'66 - SM'69 - F'87) graduated in electrical
IEEE Committee Report, "Excitation system models for
engineering from Delhi Polytechnic, India, in 1952 and obtained
power system stability studies", IEEE Trans. on PAS,
the M.E. degree from the University of Roorkee in 1962. In
Vol.PAS-100, N0.2, 1981, pp.494-509
1965 he received the Ph.D degree from the University of London
Anderson, B.D.O. and J.B. Moore, 'Linear optimal control', and D.1.C from the Imperial College, London, England. From
Prentice-Hall, Englewood Cliffs, N.J., 1971 1952 to 1961, he worked with electric utilities in India. In 1965
Lu, Q., Z.H. Wang and Y.D. Han, 'Optimal control for he worked with the English Electric Company in England. He
power transmission systems', Qinhua University Press, moved to Canada in 1966. He taught for two years at the
Beijing, 1984 (in Chinese) University of Windsors and is at present a Professor at the
Chen, G.P. and Y.H. Qin, "A sensitivity method for the University of Calgary.
weighting matrix selection of linear optimal control
systems", The Journal of Chongqing University, Vol. 11, Qin, Yi-Hong was born in Wuxi, Jiangsu, the People's
NO.3, 1988, pp.67-73 Republic of China, on May 18, 1927. He received his Diploma
in electrical engineering from Chongqing University, 1952.
[ 101 Stephen, B.H., "The generalized eigenproblem: pole-zero Presently, he is at Chongqing University and holds the rank of
computation", Proceedings of the IEEE, Vo1.76, No.2. 1988, Professor. His research interest is in the field of power system
pp. 103-120 stability analysis and control.
[ 111 Yu, Y.N., 'Electric power system dynamics', Academic
Press, INC. U.S.A., 1983 Xu, Guo-Yu is at present a Professor at Chongqing
[ 1 21 Jacobson, D.H., 'Extensions of linear-quadratic control, University. His research interest is in the field of power system
optimization and matrix theory', Academic Press, New analysis and economic load dispatch.