Download as pdf or txt
Download as pdf or txt
You are on page 1of 1

Stochastic Speed Governor based on the

Generalized Minimum Variance Controller


1 1 2
Rodrigo Trentini , Rdiger Kutzner , Lutz Hofmann
1
Hochschule Hannover, Fakultt I - Elektro- und Informationstechnik
2
Leibniz Universitt Hannover, Institut fr Elektrische Energiesysteme

Since the analysis is focused on SISO systems, a change in the voltage reference vr enters
Introduction in the system as a state disturbance.
In this work only the governor is regarded for comparisons, however any SISO linear con-
One of the simplest member of the Stochastic Control family is the Generalized Minimum
troller might be considered. Governors structure is shown in Fig. 3.
Variance Controller (GMVC). It is based on the Minimum Variance (MV) regulator and
was first introduced by Clarke and Gawthrop in 1975 [1]. GMVCs main difference to
kf
deterministic controllers is that the former is based on ARMAX (Auto-Regressive Mov-
ing Average with eXternal inputs) models, i.e. the GMVC considers a full description of
Pr Kp Ti u
systems deterministic and stochastic parts. This important feature, inherited from the -
Kalman Filter synthesis, enables the controller to distinguish between noise and uncer-
tainty, from deterministic cause and effect. Pe
Also important to remark is that, as the GMVC is a polynomial linear controller, it can
Figure 3: Turbine governor (Load control mode).
be easily written in the same structure as any other mono-variable linear controller, e.g.,
the PID controller and its derivations. It means that deterministically both GMVC and
PID may be designed to present the same output behavior in closed-loop, however the
sensitivity function of both controlled systems might differ, leading the plant controlled
by the GMVC to a more stable behavior with less influence of the measurement noise. It Deterministic vs. Stochastic controllers
is especially attractive for mechanical devices, such as governor valves, and this feature is
what is pursued in the present work. Figure 4 shows the comparison between both controllers, where the control signal vari-
(k) ance of the stochastic one is 84% lower than the result obtained through the deterministic
Plant controller.
C(q 1 )
A(q 1 )D(q 1 )

1 1
u(k) d B(q )
yr (k) T (q 1 ) z y(k) 0.94
- S(q 1 ) A(q 1 )G(q 1 )
power (P)
Output

R(q 1 ) 0.92
AVR
GMV
Figure 1: RST structure for a plant represented by a generic SISO polynomial model.
0.9
0 5 10 15 20 25 30 35 40

Control law and Closed-loop equations 0.06

The RST structure gives the following control law,


signal (u)

0.04
Control

T yr(k) R y(k)
urst(k) = , (1)
S
and then its closed-loop representation is directly obtained, 0.02
qd B T CGS
yrst(k) = d
yr(k) + d
(k). (2)
AGS + q BR AGS + q BR 0
0 5 10 15 20 25 30 35 40
On the other hand, GMVs control law is given by, time (s)
T C G yr(k) F G y(k)
ugmv(k) = , (3) Figure 4: Step of 5% in generators reference power Pr with measurement noise and state
BDE + CGQ
disturbance for kp = 0.2, Tn = 10 s and kf = 0.8. Legend: (blue) PI with feedforward;
where it becomes clear the difference to Eq. 1. However, in spite of this difference, GMVs
(red) GMVC-based governor.
closed-loop equation is,
BT CGQ + BDE
ygmv(k) = yr(k) + (k). (4)
AGQ + BP AGQ + BP
When comparing Eq. 2 and 4, one might observe that their deterministic parts are very
similar if Q(q1) = S(q1) and P(q1) = R(q1). Important to notice that Q(q1), Summary and conclusions
P(q1) and T (q1) are GMVs free tuning digital filters. The only mismatch that occurs
is due to plants delay, which is fully compensated if previously known d-steps in advance. This work has presented a framework to compare different turbine governors on a unified
On the other hand, their sensitivity functions differ due to the right hand side numer- base. By the use of the so-called RST structure it is possible to match the determinis-
ator term of ygmv, where polynominal E(q1) is obtained from the solution of GMVs tic parts of the closed loop plants of different controller types, making possible the fair
Diophantine Equation and is directly related to the noise prediction. comparison of different controller types.
The comparison framework may also be used to highlight improvement potential of con-
troller types using stochastic in their design process. It is thereby shown that a GMV
Benchmark system
controller has superior effects on the control when compared to a standard PI with feed-
The chosen benchmark system is the well-known Single-Machine Infinite-Bus System, forward with equivalent gains. It also is shown that the RST formulation allows to design
which is presented in Fig. 2. any linear SISO controller as a stochastic controller using the GMVC structure.


- References
Pr Governor u
+Turbine Pe [1] D. W. Clarke and P. J. Gawthrop, Self-tuning controller, Proceedings of the Institution of Electrical
SMIB Engineers, vol. 22, pp. 929-934, 1975.
Excitation vt
vr [2] P. Kundur, N. J. Balu and M. G. Lauby, Power system stability and control, McGraw-Hill New York,
- System
1994.
[3] IEEE Std 122-1985, Recommended Practice for Functional and Performance Characteristics of
Figure 2: Single-Machine Infinite-Bus benchmark system.
Control Systems for Steam Turbine-Generator Units, 1985.

48. Kraftwerkstechnisches Kolloquium, 18-19 Oct 2016, Dresden, GER

You might also like