You are on page 1of 6

TURBOGENERATOR SELF-TUNING AUTOMATIC VOLTAGE REGULATOR

J.W. Finch* Senior Member, IEEE

K.J. Zachariah** Member, IEEE,

*Department of Electrical and Electronic Engineering


University of Newcastle upon Tyne
Newcastle upon Tyne, NE1 7RU, UK

M. Farsi* Member, IEEE

**was with Parsons Power Generation Systems


now with Merz & McLellan,
Newcastle upon Tyne, NE4 7YQ, UK
Previous development of a digital AVR (DGAVR) had used
the same concept of a direct replacement for a conventional
analogue AVR [2]. Computer technology progress gives the
possibility of more sophisticated digital control, with similar
hardware. A new self-tuning AVR (STAVR) would adapt to
the changing operational conditions of the TG. This has the
potential to yield the best range of responses from the
combined machine/AVR system. The option of adding an
external power system stabiliser (PSS) for further
improvement, if needed, is applicable to any of these schemes.
This PSS option has also been studied. An additional benefit
of the STAVR lies in the reduction in the commissioning
process for the AVR, often done on open-circuit.

Results are presented from new self-tuning


Abstract
controllers for turbogenerators. Simulations and micro machine
tests have been successful, with a prototype self-tuning controller
being manufactured. This is intended as a direct replacement for
a conventional AVR, so a more conventional fixed parameter
controller is also considered, serving as a comparison of
performance. Typical results are given for normal set point
changes and fault dynamics. An additional modification which
could be introduced if a power signal were available, a form of
power system stabiliser, is also covered. These new controllers
are shown to have good potential for practical application.
Key words: turbogenerator control, self-tuning control, AVR

These possible benefits has meant that much attention has


been devoted to using modern control techniques for TGs,
often centred around various types of self-tuning (ST) control.
These include Minimum Variance (MV), Generalised
Minimum Variance (GMV), optimal predictor, and Pole
Assignment (PA) [1,3,4]. In much ST AVR work, additional
signals are used to improve robustness, which is avoided here
for compatibility with existing AVR designs. MV generally
gives very lively control and can be highly sensitive to nonminimum phase plant. GMV, although more robust and
generalised, is vulnerable to unknown or varying plant dead
time and can have difficulty with dc offsets. PA aims to locate
the closed-loop poles of the system at pre-specified locations
leading to 'smooth' controllers, but the algorithm can show
numerical sensitivity when the plant model is overparameterised.

I. I NTRODUCTION
The Automatic Voltage Regulator (AVR) of the
turbogenerator (TG) attempts to control the terminal voltage
and reactive power whilst also ensuring proper sharing of the
reactive power amongst parallel connected generators. TGs are
nonlinear systems which are continuously subjected to load
variations. The AVR design must cope with both normal load
and fault conditions of operation. Such operating condition
variations cause considerable changes in the system dynamics.
When conventional linear fixed-gain AVRs are used this
degrades the performance. Response variations can in some
circumstances cause the AVR to introduce negative damping,
degrading system stability [1]. The development philosophy
was to overcome these problems by developing an exact
functional replacement for a conventional AVR, with machine
voltage the only feedback signal. No extra transducers are
needed over those for a conventional AVR.

A self-tuning (ST) controller algorithm using Generalised


Predictive Control (GPC) was selected. The algorithm has
been described in the literature [5-7], so only a very brief
review of the aspects relevant to this TG controller is given
here.
In this STAVR on-line recursive parameter estimation is
employed to evaluate the time-varying or unknown parameters
of a discrete time model of the system. In the TG changes in
the system dynamics are slow and the estimator should be and
is able to track parameter variations well. The required
performance would then be given over the complete operating
range, rather than the compromise of using a fixed parameter
controller chosen at a particular operating point. Any conerns
over the ST controller ranging too far can be met in the
software by imposing limits or 'jacketing' the control.

II SELF-TUNING CONTROL

Fig. 1 shows a block diagram, and Fig. 2 a software flow


diagram of the ST and DG AVRs. The GPC algorithm is
described elsewhere [5-7].

A. Generalised Predictive Control

C. Implementation

Advances in microcomputer technology have made more


sophisticated algorithms feasible. GPC [5] was chosen here,
for these practical reasons:

Compatibility with existing hardware, and ease of upgrading


to more powerful processors were requirements.

1) Capable: can control difficult systems such as a TG


without special adjustments.

VMEbus compatible microcomputer modules centred


around a MC68030 processor were chosen for implementation
trials. A production model only requires software engineering,
trials on a full size TG will then be possible.

2) Multi-step predictor: control signal can be influenced


by future system output bounds, giving robustness.

In the developed system analogue and digital I/O are


available as required, and the terminal voltage reference can be
changed via the interface. A user can also scroll through the
various parameters of the controller for checking or modifying
purposes. 'C' was used to code the control algorithm which
uses a recursive least-squares parameter estimator with a
variable forgetting factor [8].

3) Tuning knobs: enable customised performance and


give flexibility.
4) Future target reference: can be set helping control
during scheduled changes.
5) Structure: simple or complex controllers as needed.

In practice the controller would be fed with good initial


model parameter values obtained during previous runs. These
trials used zero initial estimates, a more arduous test. Random
values of control input were therefore applied for the first 10
sample periods before initiating the control law calculation.

B. AVR Function Replacement


An important requirement of the new controllers was that
they do not require extra transducers or feed-back signals,
compared to a conventional AVR. Both the new STAVR and
the fixed parameter DGAVR operate with a single controller
input (voltage) and supply a single output (excitation signal).
Both can be direct functional replacements for a conventional
AVR. In each case auxiliary software calculates power,
VARs, and frequency. This is not part of the main controller
algorithm and is only for monitoring and protection.

Fig. 2 Contrasting flow diagram for ST and DG AVRs

Fig. 1 Block diagram for ST and DG AVR system

A conventional digital AVR which attempts the design


specification for the 'fast' excitation control system was
produced for comparisons. Such AVRs are the digital
equivalents of the sort of controller in use for many years, and
offer a good standard of performance [2]. An approximate
design using simulation studies fine-tuned by trial and error on
the micro-alternator gave the parameters of this digital AVR
as: loop gain with generator on open circuit = 325; lag time
constants = 9.0 and 0.025 seconds; lead time constant = 3.0
seconds. For future reference, this design is termed DGAVRF. The frequency response of the micro-alternator system was
obtained by a Dynamic Signal Analyser using Fast Fourier
Transforms. These tests used a small (3%) set point change in
output; the eventual field demand settles to a new steady state
value also close to a 3% change showing operation is close to
magnetically linear here. The terminal voltage response given
had overshoot 4%,, rise time 175ms, settling time 350ms and
bandwidth 2.8Hz, considered acceptably close to the design
aims.

terminal voltage, p.u.

1.03
1.02
1.01
1.00
0.0

0.5

1.0

time, s

Fig. 3 Open-circuit terminal voltage step response (micro-alternator),


STAVR-F solid line; DGAVR-F dotted line.

III. TESTING METHODS


GPC has various parameters or 'tuning knobs' which can be
chosen to vary the behaviour [5]. One such is the control
horizon Nu which specifies the number of steps over which the
demand increments are varied. Initial trials used values of 1-3,
with large values causing a faster response. Nu = 2 is a good
compromise giving a terminal voltage step response similar but
slightly improved over the previous test under identical
conditions. Values given were: overshoot 1%, rise time
180ms, settling time 300ms and bandwidth 3.0Hz. This and
the previous DGAVR-F result are shown in Fig. 3. As Nu
approaches the prediction horizon Ny (the number of steps
over which the output directly influences the controller,
typically set to 10), the step response gets closer to the design
values. The chosen value yields a reasonably 'fast' response
which is not very different from the design values, without the
possible reduction in the controller robustness and additional
computational burden imposed by higher values. For future
reference, this 'fast' design using Nu = 2 is termed STAVR-F.

A 3-phase, 4 pole micro-alternator system was used for


practical tests. The micro-alternator field is driven through a
time constant regulator; a setting of 6 seconds was used in
these tests. The DC motor drive to the micro-alternator can
also be electronically controlled to represent the turbine and its
governor, if needed. All the major system variables are
accessible for testing.
Initial tests probed controller
performance during normal operation, these were later
extended to cover behaviour with power system faults.
A specially written C code standard two-axis theory flux
linkage based state space simulation [9] allowed tests beyond
the capability of the micro-alternator system, including wide
ranging fault simulation studies. A 10th order model with
constant reactance values was used for much of the work, with
single damper coils on each d-q axis, and lumped rotor inertia.
Other model complexities are possible.

B: Step response: generator on load


IV. STEP RESPONSE TESTS

As mentioned earlier, it is the response of the excitation


control system when the turbine generator is on load that is
really important since the system operates in this mode most of
its life. The responses of the different types of AVR obtained
on open circuit in the previous section cannot normally be
achieved when the generator is on load. This is due to the
significant changes that the generator characteristics undergo
when the operating mode is changed from open circuit to the
loaded state. It was observed during an evaluation of the ST
parameter estimator that the steady state gain and dominant
time constant with load are considerably lower than their open
circuit values and the system can exhibit some degree of
oscillatory behaviour at high load conditions.

Often the specification on desired behaviour includes TG


open circuit response, this was certainly the case here.
Frequently an AVR is site tuned on open circuit.
Consequently the first tests used the micro-alternator in this
condition, at rated voltage and speed. Each controller design
was evaluated by standard tests, including applying a 3%
positive step. A small number of specifications were set as
controller design goals. The 'fast' design specification is
typical where rapid action is required for dynamic response
control. This set aims of: overshoot 4.3%, rise time 130ms,
settling time 230ms, and closed-loop system bandwidth 4.0Hz.

transient in nature are classed as abnormal operating


conditions of the generator. Although the occurrence of these
abnormal operating conditions is very infrequent, the
performance of an AVR during these events should be
evaluated to assess whether the controller is able to cope with
such situations satisfactorily.

terminal voltage, p.u.

1.10

1.08

1.06

In the case of the STAVR, the GPC cost function considers


only the deviations of the terminal voltage from its set point
and the liveliness of the control signal. However, during major
disturbances the rotor angle of the generator with respect to the
infinite busbar of the power system is disturbed significantly
and can take some time to settle down following the event. It
is generally well known that a fast acting AVR such as the ST
controller can reduce the damping torque of the generator if it
uses only the terminal voltage as its feedback signal. The
consequence of this is the reduction in the damping of rotor
oscillations following a major disturbance. This aspect should
therefore be examined in detail to ensure that sufficient
damping of rotor oscillations is provided.

1.04
0

a)

time, s

real power, p.u.

0.85
0.80
0.75
0.70
0.65
0
b)

2
time, s

The response of a turbine generator to severe disturbances


depends very much on its severity as well as the conditions of
the power system at which the disturbance occurs. A severe
disturbance, regarded as a standard test, is a 3-phase shortcircuit. The performance of the new AVR is now examined
under these conditions using the simulator.

Fig. 4 Loaded step response (micro-alternator),


a) terminal voltage, b) real power .
STAVR-F solid line; DGAVR-F dotted line.

The step response obtained when using the different AVRs


on the TG simulator, representing a typical 660MW set, has
also been investigated. A positive step of 3% was used with an
operating point of P = 0.8pu; Q = 0pu. These tests confirmed
that performance similar to that obtained with the microalternator can be repeated with the TG simulator.

A. Three phase short circuit


During this test, a sudden short circuit is applied to the
stator terminals of the generator and is removed after a period
of 100 ms The operating point of the generator has been
chosen as P = 0.8 pu, Q = 0 pu to obtain a large rotor load
angle. The greater the rotor angle the more severe is the test
since the stability margin of the rotor is lesser in that case. For
comparisons on the damping available to the rotor during the
disturbance, a factor called the 'Effective Damping Ratio'
(EDR) has been used. This factor is widely used in the
industry and is defined as the ratio of the peak-to-peak
amplitude between the first undershoot of a signal following a
disturbance and the second over-shoot to the peak-to-peak
amplitude between the first undershoot and the first overshoot.
A lower value of the EDR indicates higher damping.

Fig. 4 shows the terminal voltage step response with the


STAVR-F, values given were: overshoot 3.5%, rise time
500ms, settling time 500ms; also shown is the variation in real
power. The rise time differs markedly from OC conditions,
since the alternator systems steady state gain has changed by
about 5. The bandwidth of 3.6Hz is similar to the OC case.
The corresponding step response with the conventional
DGAVR-F (also in Fig. 4) gave overshoot 3%, rise time
840ms, settling time 840ms, bandwidth 1.7Hz, showing
considerable changes from the OC values. These results
clearly indicate that the STAVR is able to maintain its
response characteristics under changing system conditions,
while a fixed AVR fails to do so. These responses are
comparable to those on the simulator, a useful confirmation.

Fig. 5 gives the response with STAVR-F. The EDR of the


rotor angle signal is 0.64 and its settling time to within 2% is
found to be 1.25 seconds; the terminal voltage settles down in
0.39 seconds, a satisfactory performance. The test was
repeated with the conventional AVR, DGAVR-F, and a rather
similar response was obtained (also in Fig. 5). The EDR and
the settling time of the rotor angle found were 0.73 and 2.14
seconds respectively and the settling time of the terminal
voltage is 0.6 seconds. This performance indicates that the
STAVR has improved the rotor damping in this case.

V. RESPONSE TO POWER SYSTEM FAULTS


Major disturbances that occur in the power system from
time to time can seriously affect the smooth operation of the
excitation control system. These disturbances which are

terminal voltage, p.u.

terminal voltage, p.u.

1.10

1.2
1.0
0.8
0.6
0.4
0.2
0.0
1

a)

time, s

time, s
0.85

90
80

real power, p.u.

load angle, deg

1.06

1.04
0

a)

70
60
50
40

0.80
0.75
0.70
0.65

0
b)

1.08

3
b)

time, s

2
time, s

Fig. 5 3-phase SC (simulation), a) terminal voltage; b) rotor angle; ,


a) terminal voltage, b) real power .
STAVR-F solid line; DGAVR-F dotted line..

Fig. 6 Loaded step response with power signal (PSS) (micro-alternator),


a) terminal voltage, b) real power .
STAVR-F solid line; DGAVR-F dotted line.

VI. POWER SIGNAL TESTS

VI. C ONCLUSIONS

Fast acting AVRs have the potential to improve transient


behaviour of the system. A key element in the design
philosophy here was to retain the input/output replacement for
an earlier AVR, rejecting use of extra signals or transducers.
Using additional devices to improve stability has become
popular, such Power System Stabilisers (PSS) [10] may use
several auxiliary feedback signals, but power is common.

Use of GPC for the adaptive control of a turbogenerator


excitation system has been shown to offer considerable
promise. A laboratory micro-alternator and a versatile
computer simulation model were used in these studies. Both
the conventional DGAVR and the STAVR were designed
using similar performance criteria, including a requirement for
a direct input/output replacement for an existing AVR
(terminal voltage/field excitation signal).
The STAVR
performs considerably better where there is a significant
change in the system, for example between OC and loaded
conditions, but also during a severe disturbance. The good
behaviour of the new ST controller stems from it tracking the
varying dynamics of the plant and its cost function minimising
control strategy, enabling the rapid regulation of terminal
voltage. Tested disturbances used 3-phase short circuits but
have also included full load rejection, line switching, etc.

To allow comparisons, a version of the new AVR was


designed using terminal voltage, as before, and electrical
power. Step response tests on this PSS were again used to
compare behaviour.
A number of trials were done to establish an effective
balance between the terminal voltage and real power signals.
The modified step response showing both terminal voltage and
real power is shown in Fig. 6, an equivalent result with
DGAVR is included for comparison. A good standard of
behaviour is again given with this PSS STAVR, with improved
damping of power (Fig. 6b compared with Fig. 5b), but
slightly degraded voltage response (Fig. 6a compared with Fig.
5a). This is an expected consequence of the controller
attempting to control both voltage and power in the transient.

Tests have also explored the possible behavioural changes if


extra signals were employed. Use of such a power systems
stabiliser has shown improved damping of power and rotor
angle at the expense of terminal voltage. Such developments,
and the use of governor control, provide possibilities for
further developments. Meanwhile, good performance for the
use of self-tuning control for turbine generator excitation
systems has been proven.

John W. Finch (M'90, SM'92) was


born in Co. Durham, England. He
received the BSc(Eng) degree from
University College London, where he
graduated with First Class Honours in
Electrical Engineering, and the Ph.D.
from the University of Leeds. He has a
consultancy activity with many
National and International firms, and
has over 100 publications.
Areas
covered include CAD, applied control,
simulation, electrical machines and drives, robotics. He is
Reader in Electrical Control Engineering at the University of
Newcastle upon Tyne, and is an IEE Fellow, and a Chartered
Engineer. Dr Finch is a winner of the Goldsmid Medal and
Prize (UCL Faculty prize), the Carter Prize (Leeds University
post-graduate thesis prize), and the IEE's Heaviside Premium.
He serves on the IEE Professional Group P1 'Electrical
Machines and Drives', and B2 'Applied Control Techniques'.

VII. A CKNOWLEDGEMENTS
The authors thank Parsons Power Generation Systems Ltd,
the University of Newcastle upon Tyne and SERC/ERCOS for
support, the use of facilities, and for permission to publish.
VII. R EFERENCES
[1] B.W. Hogg, Representation and control of
turbogenerators in electric power systems, Chapter 5 in
Modelling of dynamical systems, Vol: 2 (H. Nicholson,
Ed), Peter Peregrinus, 1981, pp. 112-149
[2] R.S. Hingston, P.A.L. Ham, and N.J. Green,
Development of a digital excitation control system,
IEE EMDA'89 Conference, London, pp. 125-129, 1989
[3] A.S. Ibrahim, B.W. Hogg, and M.M. Sharaf, Self-tuning
automatic voltage regulator for a synchronous generator,
IEE Proc. D, 1989, 136, (5), pp. 252-260

K. J. Zachariah obtained his


BSc(Eng)
in
Electronics
&
Communications from the University
of Kerala, India in 1975, his MSc and
PhD by research in Self-tuning Control
from the University of Newcastle upon
Tyne in 1986 and 1994. From 1975 to
1984 he worked in the Kerala State
Electronics Development Corporation
Ltd., India where he specialised in the
design and engineering of industrial
systems and controllers for power plant. He joined Parsons
Power Generation Systems Ltd in 1986 where his work
involved the development of self-tuning control systems for
turbine generators; and is now with Merz & McLellan,
Newcastle upon Tyne, UK

[4] A. Chandra, K.K. Wong, O.P. Malik, and G.S. Hope,


Implementation and test results of a generalised selftuning excitation controller, IEEE Trans. En. Conv.,
1991, 6, (1), pp. 186-192
[5] D.W. Clarke, C. Mohtadi, and P.S. Tuffs, 1987,
Generalised predictive control, the basic algorithm, and
extensions and interpretations, Automatica, 23, (2), pp.
137-160
[6] M. Farsi, K.J. Zachariah, J.W. Finch, and P.A.L. Ham,
A self-tuning regulator for turbogenerators, ACC'91,
Boston, pp. 1026-31, 1991
[7] J.W. Finch, K.J. Zachariah, and M. Farsi, Self-tuning
control applied to turbogenerator AVRs, IEE Proc.
GTD, Sept. 1996, 143, (5), pp. 492-499

Mohammad Farsi was born in


Shahriar, Tehran, Iran in 1940. He
received his BSc degree in Electrical
Engineering from the University of
Idaho in 1968. He worked as Weapon
Electrical Officer in the Iranian Navy
up to 1982. He received his PhD in
Self-tuning Control and Model
Reduction in Robotics from the
University of Newcastle upon Tyne in
1986; where he is now Senior Lecturer.
His research interests lie in optimal, adaptive, intelligent,
neural and genetic algorithm controllers, with applications in
robotics and automation. He has over 80 publications and
serves on the IEE Professional Group 'Applied Control
Techniques', C9.

[8] T.R. Fortescue., L.S. Kershenbaum, B.E. Ydstie,


Implementation of self-tuning regulators with variable
forgetting factors, Automatica, 17, 6, pp 831-835, 1981
[9] B. Adkins, and R.G. Harley, The general theory of
alternating current machines, Chapman and Hall,
London, 1975
[10] N.C. Pahalawathatha, G.S. Hope, and O.P. Malik,
MIMO self-tuning power system stabiliser, Int. J.
Control, 1991, 54, (4), pp. 815-829

You might also like