You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3937019

Small signal analysis of hydro-turbine governors in large interconnected power


plants

Conference Paper · February 2002


DOI: 10.1109/PESW.2002.985195 · Source: IEEE Xplore

CITATIONS READS
28 2,516

3 authors:

Innocent Kamwa D. Lefebvre


Laval University 21 PUBLICATIONS   490 CITATIONS   
420 PUBLICATIONS   12,714 CITATIONS   
SEE PROFILE
SEE PROFILE

Lester William Loud


Hydro-Québec
15 PUBLICATIONS   265 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

AI-Driven Hybrid Renewable Energy Systems View project

Phasor Measurement for Wide Area Protection & Control View project

All content following this page was uploaded by Innocent Kamwa on 13 May 2014.

The user has requested enhancement of the downloaded file.


Small Signal Analysis of Hydro-Turbine
Governors in Large Interconnected
Power Plants
Innocent Kamwa, Senior Member, IEEE, Daniel Lefebvre, Lester Loud, Member, IEEE

Abstract—The paper describes a transient stability program However, improved tuning of hydro-turbine governor,
(TSP) based approach to identify numerically, a state-space, especially for grid-connected unit can be easily achieved
small–signal model of the open-loop system seen by the hydro- through a small-signal modeling of the external network, seen
turbine governor during its normal operation. This single-input
as the process to be controlled. To this aim, we are proposing a
multiple-output model is validated successfully by comparing
actual closed-loop responses computed in the TSP with those system identification approach for determining from pulse
simulated in Matlab using the linearized model. For illustration, response signals simulated in the TSP, a state-space model
three governing systems found in the Hydro-Quebec’s grid are which can serve as a basis for system performance analysis and
studied: (1) a mechanical hydraulic; (2) a Woodward PID and (3) optimization. As all linearization methods, this one is
a classical Neyrpic’s governor. System performance with respect rigorously valid for the considered operating point only, as
to the speed of response, interarea modes sensitivity and closed-
discussed in [2]. However, compared with a linearization by
loop gain and phase margins is assessed, evidencing the somewhat
detrimental impact of the derivative control term for direct or numerical differentiation, the small-excitation based
interconnected operation. Overall, the small-signal analysis system identification [3-5] has the significant advantage of
approach devised appears well-suited for determining and/or using the same modeling engine as the utility’s TSP, without
assessing unit-connected settings of hydro-turbine speed governing the need for a separate database of models which has to be
systems. supported on its own.
The proposed approach will be illustrated with three typical
Index Terms—Speed governors, small-signal analysis, state- speed governors: a mechanical hydraulic, a Woodward PID and
space system identification.
a classical Neyrpic PD governor, found at the Hydro-Quebec’s
Carillon, Churchill Falls and Robert-Bourassa powerhouses
I. INTRODUCTION
respectively. It will be shown that an open-loop system

I N recent power systems incidents, a poor tuning of hydro-


turbine governors has proven to be a detrimental factor in the
amplification and propagation of the disasters. The fact that in
identification based on small-signal signal responses to a one-
second pulse applied on the gate opening yields a highly
accurate linearized representation of the interconnected grid
simulations, damping of low-frequency interarea modes is often which from the speed governing standpoint, can be
significantly improved by blocking the hydro-turbine instrumental in refining the controller parameters using modern
governors, is yet another confirmation that their parameters are model-based control theory [6].
not set to help the interconnected network during major
disturbances [1]. Part of the problem resides in the requirement II. SMALL SIGNAL SPACE MODEL IDENTIFICATION
for plant operators to keep track of two different sets of
governor parameters: one set for isolated mode of operation and A. System description
the other when connected to the grid. Since these two settings For this study, we will consider a year 2001 standard Hydro-
have conflicting goals, any confusion can result in a Quebec’s grid modeled with 784 busses, 1072 lines, 130
problematic behavior when the machine is running dispatched generators and 9 synchronous condensers, 3 HVDCs, 10
by the control center. SVCs, non-linear loads, on-load tap changers and the Hydro-
Québec’s defense plan against extreme contingencies. All the
machines are modeled using a 5th-order representation
I. Kamwa and L. Loud are with Hydro-Quebec/IREQ, Power System including saturation, while the hydro-turbines have detailed
Analysis, Operation and Control, 1800 Lionel-Boulet, Varennes (Qc) Canada, nonlinear models with parameters assessed by field tests.
J3X 1S1 (e-mail: kamwa@ireq.ca). However, only one load-flow is considered (winter peak load),
D. Lefebvre is with Hydro-Québec, TransÉnergie/Operations Planning,
Complexe Desjardins, Montréal, Québec, Canada H5B 1H7 this being an exploratory type study. Three hydro-powerhouses
. are chosen based on their relative importance for the network
and the type of speed governor they are equipped with.

0-7803-7322-7/02/$17.00 © 2002 IEEE


1178
0-7803-7322-7/02/$17.00 (c) 2002 IEEE
The first powerhouse is Carillon, near Montreal. It is a 14-units analysis results. On the same picture, the frequency response
plant commissioned in 1962, equipped with rotating exciters magnitude shows that the Hydro-Quebec’s network, which is
and mechanical hydraulic governors. The whole plant is basically a standalone power system asynchronously connected
modeled as a single 770-MVA equivalent generator. The to neighbors systems through DC lines, has a low natural
second powerhouse is Churchill Falls, a 5500-MVA installation oscillation frequency around 0.06Hz. In addition, any plant has
commissioned in 1971 and located in the far north-east of the as should be expected its own local modes, while the main
system. Its 11-units are equipped with static exciters and interarea mode located around 0.7Hz involves Churchill and
Woodward PID governors. In the TSP, this plant is modeled Robert-Bourassa together.
with three subsets machines: four units totaling 2000-MVA Time Response

(Machine No. 60); one unit rated at 500MVA (No. 79) and six
10 Frequency Resp. Peak = 0.10866 at fn = 1.242Hz;
amp = 0.016196; fn = 0.061028Hz; zeta = 0.32848
amp = 0.0064587; fn = 1.2465Hz; zeta = 0.061407
other units totaling 3000-MVA (No. 59). Our present study 5

Speed(mHz)
concentrates on machine 60. The last studied powerhouse is 0

Robert-Bourassa (previously known as LG2), a 5900-MVA −5

installation commissioned in 1981 and located in the far north- −10

west of the system. Its 16-units are equipped with static exciters 0 5 10 15 20 25
seconds
30 35 40 45 50

and Neyrpic classical PD electronic governors. The whole Bode Magnitude Response

plant is divided in three machine groups: seven units totaling 0.1

2590-MVA (Machine No. 49); one unit rated at 370MVA (No 0.08

gain(pu/pu)
449) and eight other units totaling 2960-MVA (No. 549). This
0.06

study focuses on machine 49.


0.04

0.02

B. System Identification 10
−2
10
−1
10
0
10
1

(a)
For each plant, the test setup as implemented in the TSP is
Hz

Time Response
sketched in Fig.1. The “OpenLoopFlag” is used to control a
switch which flips from an open-loop test mode (with governor 0.5

bypassed) to a normal closed-loop operation, with the unit


Speed(mHz)

connected to the grid. It should be observed that the nonlinear 0


Frequency Resp. Peak = 0.0061274 at fn = 1.242Hz;
hydro-turbine as well as the excitation system are considered to −0.5
amp = 0.0017496; fn = 0.058336Hz; zeta = 0.3162
amp = 0.00099203; fn = 1.2148Hz; zeta = 0.14905

be part of the large external system. However, the acceleration amp = 0.00084769; fn = 1.9104Hz; zeta = 0.23549
amp = 0.00044964; fn = 1.6106Hz; zeta = 0.22962
amp = 0.00027883; fn = 0.72016Hz; zeta = 0.14594
power-based PSSs present on units 49 and 60 are deactivated 0 5 10 15 20 25
seconds
30 35 40 45 50

prior to open-loop tests in order to simplify the subsequent −3 Bode Magnitude Response
x 10
control system studies in Matlab. 6

4
gain(pu/pu)

Machine
3
Under Test External Network
Pulse/Step 2
Excitation A Va
hydraulic
OpenLoopFlag Pm 1
turbine B Vb
−2 −1
Switch C Vc 10 10 10
0
10
1

wref wref Vf Hz (b)


m Time Response
we
gate opening
4
Pref Pref
Field Voltage vqd vs_qd 2
Pe
Speed(mHz)

0
Neyrpic/Woodward Excitation System wm m
Governor −2 Frequency Resp. Peak = 0.053727 at fn = 0.055103Hz;
amp = 0.023631; fn = 0.057977Hz; zeta = 0.33048
Pe −4 amp = 0.0050049; fn = 0.95914Hz; zeta = 0.31082
amp = 0.0038804; fn = 1.5384Hz; zeta = 0.1684
amp = 0.0018622; fn = 0.72649Hz; zeta = 0.16036
Machines Measurement −6
amp = 0.00099663; fn = 1.3768Hz; zeta = 0.1385
0 5 10 15 20 25 30 35 40 45 50
seconds

Fig.1. Test setup in the transient stability program (TSP) Bode Magnitude Response

0.05
The probing signal is a one-second pulse with a 1% magnitude. 0.04
For each plant, six signal responses are selected for developing
gain(pu/pu)

0.03
a single input multiple output model, including the unit under
0.02
test’s speed and electrical power. From the goodness of fit
perspective, the resulting 40th -order state-space model is nearly
0.01

perfect, with the actual TSP based signals practically 10


−2
10
−1 0
10
1
10
(c) Hz
undistinguishable from the linear model responses. Figure 2 Fig.2. Model-based open-loop speed response to a 1% and one-second pulse
illustrates the small-signal speed responses simulated with the excitation applied to the gate opening. (a) Carillon (#7); (b) Churchill(#60);
identified open-loop model, along with an excerpt of the modal (c) Robert_Bourassa(#49). Only the dominant modes are listed in the caption.

1179
0-7803-7322-7/02/$17.00 (c) 2002 IEEE
III. HYDRO-TURBINE GOVERNOR MODELS IN MATLAB wref
1
In order to fully take advantage of the small-signal model just
derived in control system performance analysis, we found it 2 kd.s
suitable to first implement the target speed governing systems we td.s+1
in Matlab/Simulink (Figs 3-5). Every governor presented 3 Permanent
Derivative
droop
consists of three main parts: Pref
Rp
4
• the servomotor subsystem including an integrator and
Pe
a gain: its parameters are fixed, and manufacturer
dependant servomotor speed limit
1 1
• the permanent droop whose value is fixed at Rp=5% in ka
s
1
ta.s+1
the Hydro-Québec’s system. Notice that the droop gate
opening
signal is plant dependant and can be the opening gate Temporary Droop

(Fig. 3) or electrical power (Figs. 4 and 5). beta*tw.s


tw.s+1
• the adjustable controller which in its simplest
expression, reduces to a PID (Fig. 4). In Fig. 3, the Fig.5. Neyrpic Electronic Governor.
Ta=0.05s;Ka=5;kd=0.8;td=0.2s;beta=0.5;tw=5s;
only adjustable controller block is the temporary droop Electrical power transducer time constant: 0.44s
while in Fig. 5, a derivative section can be tuned to Frequency transducer time constant: 0.08s
supplement the temporary droop. Controllers Frequency Responses
Although transducers time constants are shown under Figs
3-5, their dynamical impact is already included in the open- 8
Woodward(Churchill)
Mechanical(Carillon)
loop model. Figures 6 and 7 illustrate the frequency-domain Neyrpic(Robert−Bourassa)

Gain(p.u.)
6
behavior of the various governors assuming the default settings
built into the Hydro-Quebec’s transient stability program. 4

2
wref 1
speed
servomotor limit −2 −1 0 1
10 10 10 10
1 1
ka 1
2 s
ta.s+1
we gate
opening 0
Temporary Droop
−50
beta*tw.s
Phase(Deg.)

tw.s+1 −100

Rp −150

permanent droop −200

−2 −1 0 1
10 10 10
Fig. 3. Mechanical Hydraulic Governor. Hz
10

Ta=0;Ka=3.333s;Rp=0.05;beta=0.3;tw=3s. Fig.6. Frequency responses of the governors


Frequency transducer time constant: 0.8s
Controller sensitivities
kp
3
1.15
wref Proportional
1.1
4 1
ki
s
S(p.u.)

1.05
we
Integral
1

1 droop kd.s
Rp 0.95

Pref td.s+1 0.9


Derivative
2 10
−2 −1
10 10
0
10
1

Pe

speed 0.4 Woodward(Churchill)


servo1 position servo2
limit 0.35 Mechanical(Carillon)
ka 1 1 Neyrpic(Robert−Bourassa)
Ksm 1 0.3
T C.s+1 s ta.s+1
gate 0.25
T(p.u.)

opening 0.2
0.15
0.1

Fig. 4. Woodward Analog PID Governor 0.05

Ka=1;Ta=0.04s; Ksm=4;TC=0.1;Rp=0.05; kp=3.2;ki=0.45;kd=1.5;td=0.02s; 10


−2 −1
10
0
10 10
1

Electrical Power transducer time constant: 1.9s Fig.7. Frequency sensitivity of the governors with respect to output (S) and
Frequency transducer time constant: 0.03s input (T) noise

1180
0-7803-7322-7/02/$17.00 (c) 2002 IEEE
The interesting point regarding the S-function in Fig. 7 is that 10

the mechanical governor is much less sensitive than the two 5

Speed(mHz)
others. Hence, the Churchill’s PID governor is particularly
0
sensitive to output disturbance around local mode while the
Robert-Bourassa’s Neyrpic device is more sensitive to −5

interarea disturbance around 0.2Hz. −10


0 5 10 15 20 25 30 35 40 45

IV. CLOSED-LOOP MODEL VALIDATION x 10


−4

A convincing way to demonstrate the adequacy of an identified 1.5


2 Actual Closed−Loop
Linear Model Based Closed−Loop
model, especially with regards to control applications, is to

Gate Opening(p.u.)
1
compare its closed loop with those directly measured on the 0.5

system [3,4]. To this aim, closed-loop one-second duration 0

pulse responses were first simulated in the TSP using the built- −0.5
−1
in settings of the various governors. Then, the state-space −1.5

open-loop model identified in section II was installed in 0 5 10 15 20 25


seconds
30 35 40 45
(a)
closed-loop with the Simulink-based governors models in 30

Section III. The overall configuration for the Matlab closed- 20

loop simulation is shown in Fig. 8. A switch controlled by the

Speed(mHz)
10

“ClosedLoopFlag” allows to flips into the test mode, where a 0

pulse or a step is injected at the frequency error summing −10

−20
point. Notice that this is in contrast with the open-loop model −30
identification test, where the probing signal was injected at the 0 5 10 15 20 25 30 35 40 45

gate opening.
0.01 Actual Closed−Loop
Linear Model Based Closed−Loop
0.005
Power(p.u.)

Pulse/Step Neyrpic/Woodward
0
Excitation Governor
−0.005
wref
ClosedLoopTest
Switch −0.01
we x' = Ax+Bu
wref gate opening
Pref Pref y = Cx+Du 0 5 10 15 20 25 30 35 40 45
seconds (b)
Pe
Turbine + Machine Under Test 15
+ AVR + External Network 10
Speed(mHz)

wm 5
m
Pe 0

−5
Transducers
−10

0 5 10 15 20 25 30 35 40 45
Fig.8. Closed-loop test configuration in Matlab.

Figure 9 confirms that all identified models are satisfactory 0.03


Actual Closed−Loop
Linear Model Based Closed−Loop
0.02
from the standpoint of replicating the closed-loop behavior, 0.01
irrespective of the hydro-governor or hydro-turbine type. Since
Power(p.u.)

the small-signal state-space model built from TSP simulations −0.01

allows to closely mimic the closed-loop response of the large −0.02


−0.03
interconnected system, it can safely be used in Matlab (i.e. −0.04
outside the TSP) to analyze and tune the governor performance 0 5 10 15 20 25
seconds
30 35
(c)
40 45

when the system is dispatched and on-line. Fig.9. Model cross-validation using closed-loop signal responses in the TSP
and Matlab.
V. CONTROL LOOP PERFORMANCE ASSESSMENT Table 1: performance measures of the governing systems for the standard
Table 1 summarizes the performance of the governing system and modified (derivative-free) parameters sets
Performance Mechanical Woodward Neyrpic
at the three chosen plants. For the Woodward and Neyrpic Measure Carillon Churchill Falls Robert-Bourassa
governors, a modified setting was derived from the standard standard modified standard modified
settings by eliminating the derivative control term, i.e. by 66%-response time 37.8s 46.27s 46.18s 32.38s 33.07s
adopting kd=0 with all other parameters left unchanged. Since 95%-response time 95.92s 127.7s 127.7s 79.22s 79.53s
the governor is a two-input system, conventional gain and Phase margin 51.95 o 44.5o 50.1o 42.6o 49.7 o
Gain margin 18.2dB 12.3dB 16.3dB 11.2dB 16dB
phase margins are not well-defined. Therefore, the values in Frequency dip for N/A 64.0 68.5 72.8 73.7
Table 1 were derived using the MIMO-based definitions given a 1%-step (mHz)
in [6].

1181
0-7803-7322-7/02/$17.00 (c) 2002 IEEE
Furthermore, the frequency dip is simply the first overshoot and 95% are basically not sensitive to this derivative term (Fig.
above the steady-state level illustrated in Figs. 10-11. 11). As illustrated by Fig. 10, the most significant difference
However, since mechanical governor step response is not between standard and modified settings is the bigger frequency
oscillatory, such a quantity is undefined for that case. sensitivity of the former, characterized by a very oscillatory
Interestingly, all the governors settings are rather robust, from initial time response, and large peaks of Bode magnitude
the standpoint of MIMO gain and phase margins. In addition, response in the interarea and local frequency bands. This
removing the derivative control have no evident detrimental lower frequency sensitivity in fact explains the better
impact on the overall performance measures. However, being robustness margins shown in Table 1 for derivative-free
based on an open-loop model with the unit connected to the governors.
grid, the measures in Table 1 tell nothing about the settings
performance during an isolated performance.
Time Response
0.18

0.16

Time Response 0.14


Modified
Modified Standard
10
Standard 0.12

0 0.1

Power (pu)
−10 0.08

0.06
Speed (mHz)

−20
0.04
−30
0.02
−40
0
−50
−0.02
−60
0 50 100 150
−70 seconds
0 5 10 15 20 25 30 35 40 45 50
seconds

Bode Magnitude Response

Bode Magnitude Response

7
0.18

0.16 6

0.14
5
0.12
Power (pu)
Speed (pu/pu)

0.1 4

0.08
3
0.06
2
0.04

0.02 1

−2 −1 0 1

(a)
10 10 10 10
−2 −1 0 1
Hz 10 10 10 10
Time Response Hz (a)
Modified
Time Response
60
Standard

40 0.16

20 0.14
Speed (mHz)

0 0.12

−20 0.1
Power (pu)

Modified
Standard
−40 0.08

−60 0.06

−80 0.04

0 5 10 15 20 25 30 35 40 45 50 0.02
seconds

Bode Magnitude Response 0 50 100 150


seconds
1.1

0.9
Bode Magnitude Response
0.8
Speed (pu/pu)

0.7
6
0.6

0.5
5
0.4

0.3

0.2
4
Power (pu)

0.1

3
−2 −1 0 1

(b)
10 10 10 10
Hz

Fig.10. Impact of the derivative control term on speed responses to a 1% 2

frequency error step: standard vs modified settings. (a) Robert-Bourassa


(#49); (b) Churchill Falls (#60) 1

For both the Woodward and Neyrpic governors, the robustness 10


−2 −1
10
Hz
10
0

(b)
10
1

margins improved significantly with kd=0, while the frequency Fig.11. Impact of the derivative control term on electrical power responses
dip is deteriorated, only slightly for the Neyrpic’s but more to a 1% frequency error step: standard vs modified settings. (a) Robert-
significantly for the Woodward’s device. Response time at 66 Bourassa (#49); (b) Churchill Falls (#60)

1182
0-7803-7322-7/02/$17.00 (c) 2002 IEEE
VI. CONCLUSIONS [3] I. Kamwa, L. Gérin-Lajoie, "State-Space System Identification —
Towards MIMO Models for Modal Analysis and Optimization of Bulk
This paper set forth a practical approach for deriving in a fast Power Systems," IEEE Trans. on Power Syst. , 15(1), February 2000, pp.
and easy way, effective small signal state-space models for 326-335
analyzing and tuning hydro-turbine speed governing systems, [4] J.R. Smith, J.F. Hauer, D.J. Trudnowski, F. Fatehi, C.S. Wood, "Transfer
Function Identification in Power System Applications," IEEE Trans. on
especially when the unit is dispatched and on-line. Small level Power Syst., 8(3), August 1993, pp.1282-1290.
one-second duration pulse are injected at the opening gate [5] J.J. Sanchez-Gasca, K. Clark, N.W. Miller, H. Okamoto, A. Kurita, J.H.
junction and the signal responses of the unit simulated with a Chow, “Identifying Linear Models from Time Domain Simulations,”
TSP and fitted to a black-box model using a powerful state- IEEE Computer App. in Power, April 1997, pp.26-30.
[6] I. Kamwa, L. Trudel, L. Gérin-Lajoie, "Robust Design and Coordination
space identification technique [3]. It is demonstrated on the of Multiple Damping Controllers Using Constrained Optimization," IEEE
Hydro-Quebec’s year 2001 network that closed-loop behavior Trans. on Power Systems, 15(3), pp. 1084-1092, August 2000
of the mechanical, Woodward PID and Neyrpic PD governors
installed at the Carillon, Churchill Falls and Robert-Bourassa VIII. BIOGRAPHIES
powerhouses respectively, can be satisfactorily replicated in
Matlab based on the identified 40th-order open-loop plant Innocent Kamwa (S'83, M'88, SM’98) has been with the Hydro-Québec
models. The later are then used with a good confidence to research institute, IREQ, since 1988. At present, he is a Senior Researcher in
assess the performance and robustness of the existing the Power System Analysis, Operation and Control Department. He is also an
associate professor of Electrical Engineering at Laval University in Québec,
parameters and provide original results illustrating the Canada. Kamwa received his BEng. and PhD Degrees in Electrical Engineering
potentially detrimental impact of the derivative control term from Laval University in 1984 and 1988 respectively. A member of the IEEE
when the generator is interconnected with the grid. Power Engineering and Control System societies, Kamwa is a registered
professional engineer.
VII. REFERENCES Daniel Lefebvre received his BEng in Electrical Engineering from École
Polytechnique, Montréal, in 1985. Since then he has been with Hydro-Quebec,
[1] D. Kosterev, B. Mittelstadt, M. Viles, B. Truck, J. Burns, M. Kwok, KJ. TransEnergie Division where he is involved in operations planning. He is a
Jardim, G. Garnett, “Model Validation and Analysis of WSCC System registered professional engineer.
Oscillations Following Alberta Separation on August 4, 2000,” published
at : http://www.transmission.bpa.gov/orgs/opi/Wide_Area/August4Report.doc. Lester Loud (M’97) received, in Electrical Engineering, his BEng and
[2] L.N. Hannet, B. Fardanesh, J.W. Feltes, "Field Tests to Validate Hydro MEng from Concordia University in 1985 and 1988 respectively and his PhD
Turbine-Governor Model Structuire and Parameters,” IEEE Trans. on from McGill University in 1996. He has been with Hydro-Québec research
Power Syst. , 9(4), Nov. 1994, pp. 1744-1751. institute, IREQ, since 1994.

1183
0-7803-7322-7/02/$17.00 (c) 2002 IEEE
View publication stats

You might also like