EXCITATION CONTROL OF

A TURBO- GENERATOR BY
USING
ARTIFICIAL NEURAL
NETWORKS

G. Vishnu Vardhan
[email protected]

S. BrahmaNanda Reddy
[email protected]

IV B.TECH

ELECTRICAL AND ELECTRONICS ENGINEERING

Malineni Lakshmaiah Engineering College,

ABSTRACT:
The critical factor affecting the modern power system today is load flow control.
This load flow control is affected by the excitation and frequency (speed) of the
machines in the system. This paper presents the design of two separate continually
Online trained (COT) neurocontrollers for excitation and turbine control of a turbo
generator connected to the infinite bus through a transmission line.
These neurocontrollers are augment/replace the conventional automatic
voltage regulator and the turbine governor of a generator. A third COT artificial
neural network is used to identify the complex nonlinear dynamics of the power
system. Results are presented to show that the two COT neurocontrollers can
control turbo generators under steady state as well as transient conditions and, thus,
allow turbo generators to operate more closely to their steady-state stability limits.
INTRODUCTION:
TURBOGENERATORS supply most of the electrical energy produced
by mankind and, therefore, form major components in electric power systems and their
performance is directly related to security and stability of power system operation. A
turbo generator is a nonlinear, fast-acting, multivariable system, and is usually connected
through a transmission system to the rest of the power system. Their dynamic
characteristics vary as conditions change, but the outputs have to be coordinated so as to
satisfy the requirements of power system operation. Conventional automatic voltage
regulators (AVRs) and turbine governors are designed to control, in some optimal
fashion, the turbo generator around one operating point; at any other point the generator’s
performance is degraded.
Various techniques have been developed to design generic controllers
for unknown turbo generator systems. Most adaptive control algorithms use linear
models, with certain assumptions of types of noise and possible disturbances. The turbo
generator system is nonlinear, with complex dynamic and transient processes; hence, it
cannot be completely described by such linear models.
With the issues of unmodeled dynamics and robustness arise in practical
applications of these adaptive control algorithms and, hence, supervisory control is
required.

Artificial neural networks (ANNs) offer an alternative for generic controllers. They are
good at identifying and controlling nonlinear systems. They are suitable for multivariable
applications, where they can easily identify the interactions between the inputs and outputs.
It has been shown that a multiplayer feed forward neural network using deviation signals
(for example, deviation of terminal voltage from its steady value) as inputs can identify the
complex and nonlinear dynamics of a single machine infinite bus configuration with
sufficient accuracy to then be used to design a generic controller which yields optimal
dynamic system response irrespective of the load and system configurations.
An Artificial Neural Network (ANN) is an information-processing paradigm
that is inspired by the way biological nervous systems, such as the brain, process
information. The key element of this paradigm is the novel structure of the information
processing system. It is composed of a large number of highly interconnected processing
elements (neurons) working in unison to solve specific problems. ANNs, like people, learn
by example. An ANN is configured for a specific application, such as pattern recognition
or data classification, through a learning process.
Historical background:

Neural network simulations appear to be a recent development. However, this

field was established before the advent of computers, and has survived at least one major
setback and several eras.

Many import and advances have been boosted by the use of inexpensive computer
emulations. Following an initial period of enthusiasm, the field survived a period of
frustration and disrepute. During this period when funding and professional support was
minimal, relatively few researchers made important advances. Currently, the neural
network field enjoys a resurgence of interest and a corresponding increase in funding.

The first artificial neuron was produced in 1943 by the neurophysiologist Warren
McCulloch and the logician Walter Pits. But the technology available at that time did not
allow them to do too much.

Reasons for using Neural Networks:

Either humans or other computer techniques can use neural networks,

with their remarkable ability to derive meaning from complicated or imprecise
data, to extract patterns and detect trends that are too complex to be noticed. A
trained neural network can be thought of as an "expert" in the category of
information it has been given to analyze
Human and Artificial Neurons :
Learning process of Human Brain :

In the human brain, a typical neuron collects signals from others through a host of
fine structures called dendrites. The neuron sends out spikes of electrical activity through
a long, thin stand known as an axon, which splits into thousands of branches. At the end
of each branch, a structure called a synapse converts the activity from the axon into
electrical effects that inhibit or excite activity from the axon into electrical effects that
inhibit or excite activity in the connected Neurons. When a neuron receives excitatory
input that is sufficiently large compared with its inhibitory input, it sends a spike of
electrical activity down its axon. Learning occurs by changing the effectiveness of the
synapses so that the influence of one neuron on another changes

Components of a Neuron

The Synapse

This paper presents a new design and implementation of two separate

COT neurocontrollers on a single turbo generator infinite bus system; one ANN
controls the excitation and the other ANN controls the steam into the turbine with
different sampling rates. In particular, the paper makes the following new
contributions.

 It shows that the two smaller neurocontrollers achieve slightly better

performance than with the single combined neurocontroller for a wide range
of operating conditions and system configurations.
  As a consequence, it is possible to reduce the computational demand and
learning time of the neurocontrollers for real-time implementation in this
manner.

SINGLE-MACHINE INFINITE BUS

SYSTEM:
The COT neurocontrollers are designed for and evaluated by
simulation on a specially instrumented 3-kW micro alternator with per-unit parameters
typical of those expected of 30–1000 MW generators. It is also equipped with a

traditional governor and excitation controls connected to an infinite busUm, through a

transmission line, as shown in Fig. 1. A specially controlled dc motor acting as a turbine
simulator drives the micro alternator.

FIG:1 SINGLE MACHINE INFINITE BUS CONFIGURATION

The nonlinear time-invariant system equations for the system in Fig. 1 are of the form

(1)

Where g (x) contains the nonlinear terms. Equation (1) is developed from the
synchronous machine
Equations with the following selected states:

(2)
Where the first two states are the rotor angle and the speed deviation, the other states are

the currents in the d, q, field, and damper coils. The conventional AVR and excitation
system are modeled in state space as a second-order device with limits on its output
Voltage levels. The turbine simulator and governor system are modeled in state space as
a fourth-order device so that reheating between the high-pressure and intermediate-
pressure stages may be included in the model. The output of the turbine simulator is
limited between 0%–120%.

NEURO CONTROLLERS:
The ability of neural networks to model nonlinear dynamical systems has led to the
development of numerous neural network- based control strategies. Most of these
strategies are simply nonlinear extensions of existing linear techniques, such as direct
inverse control, model reference adaptive control, predictive control, and internal model
control. There are a number of successful applications of such ANN based controllers

FIG 2: SINGLE-MACHINE INFINITE BUS CONFIGURATION WITH TWO

SEPARATE NEUROCONTROLLER

(also called neurocontrollers). However, there are still many unresolved issues relating to
their use. Stability and robustness cannot be guaranteed in general for most ANN based
controllers, especially if the ANN appears directly in the Control/feedback loop. This is
because the mathematical framework for dealing with nonlinear control techniques has
not yet been developed. The single-machine infinite bus system with the ANN identifier
and the two neurocontrollers is shown in Fig. 2. This paper presents results with two
separate neurocontrollers that are trained using different sampling frequencies as shown
in Fig. 3. The ANN identifier is pretrained before the neurocontrollers’ training starts.
 FIG 3:TWO SEPARATE NEUROCONTROLLER ARCHITECTURE
The two neurocontrollers are trained simultaneously. The operation of the architecture
shown in Fig. 3 is summarized as follows.
1) The terminal voltage deviation and speed deviation signals from their set points for the
turbo generator are sampled at D and time delayed.
2) The sampled signals from step 1) are input at A to the excitation neurocontroller, and
turbine neurocontroller and these controllers calculate the damping signals for the turbo
generator.
3) The damping signals from step 2) are input at B to the turbo generator and the same
damping signals plus the signals from step 1) are input to the ANN identifier at C.
4) The output of the turbo generator at D and ANN identifier at E are subtracted to
produce a first error signal F which, via back propagation at G, is used to update the
weights in the ANN identifier.
5) Steps 2) and 3) are now repeated using the same signal values obtained in step 1), with
the ANN identifier weights fixed, and the output of the ANN identifier at E, and the
desired output at M, are subtracted to produce a second error signal at H.
6) The error signal from step 5) is back propagated at I through theANNidentifier J and K
obtained at and with the fixed weights in the ANN identifier.
7) The back propagated signals, J and K from step 6) are subtracted from the output
signals of the excitation and turbine neurocontrollers, respectively, to produce error
signals L and N.
8) The error signals at L and N from step 7) are used to update the weights in the
neurocontrollers, using the back propagation algorithm.
9) New control signals are calculated using the updated weights in step 8) and are applied
to the turbo generator at B again, to provide the required damping.
10) Steps 1)–9) are repeated for all subsequent time periods.
The ANN identifier in Fig. 2 is required to produce the error signals J and K,
which are used to update the weights in the neurocontrollers. With the use of this ANN
identifier, the need to know the turbo generator Jacobian is avoided. Also, with the use of
the ANN identifier, the neurocontrollers become adaptive and, thus, accurately control
the turbo generator under all operating conditions.
A. ANN Identifier Architecture
The ANN identifier structure is fixed as a three-layer feed forward neural network with
12 inputs, a single hidden layer with 14 neurons, and two outputs. The inputs are the
actual deviation in the input to the exciter, the actual deviation in the input to the turbine,
the actual terminal voltage deviation and the actual speed deviation of the generator.
These four inputs are time delayed and together with the eight previously delayed values
form the 12 inputs for the model. The ANN model outputs are the estimated terminal
voltage deviation and estimated speed deviation of the turbo generator.
B. Neurocontroller Architecture
The inputs to the excitation neurocontroller are time delayed by 20 ms and those to the
turbine neurocontroller are time delayed by 100 ms. The reason for the choice of a slower
sampling period for the turbine neurocontroller is because of slower response of the
mechanical system due to its inertia.
C. Desired Response Predictor
The desired response predictor is designed to have the following characteristics.
1) It must be flexible enough to modify the performance of the turbo generator.
2) The desired response signal at must ensure that the turbo generator is inherently stable
at all times. In other words, the predictor must be stable.
3) The desired response signal must incorporate the effects of a power system stabilizer.

RESULTS:
Use of Two Separate Neurocontrollers:
The dynamic and transient operation of the neurocontrollers are compared with the
operation of the conventional controller (AVR and turbine governor) under two
different conditions: +/-5% step changes in the terminal voltage set point and a
temporary three-phase short circuit on the infinite bus. The performance of the two
neurocontrollers in Fig. 2 (switches S1 and S2 in position “b”) is compared with that
of the conventional AVR and governor controllers (switches S1 and S2 in position
“a”) by evaluating how quickly they respond and damp out oscillations in the
terminal voltage and rotor angle. Restoring terminal voltage and rotor angle to
steady state after any changes is important for the stability of the power system.

1) Step Changes in the Terminal Voltage Reference VREF or Ve (Fig. 2):

Figs. 4 and 5 show the terminal voltage and the rotor angle of the turbo generator for
5% step changes in the terminal voltage with the turbo generator operating at 1 pu
power and 0.85 lagging power factor, and line impedance Z=0.02+J0.4 pu. The
neurocontrollers clearly outperform the conventional controllers.

Fig. 5. Rotor angle for +/-5% step

Fig. 4. +/- 5% step change in the
change in the desired terminal voltage
desired terminal voltage (P = 1 pu
(P =1 pu and pf = 0.85 lagging).
and pf =0.85 lagging).

2) Step Changes in the Terminal Voltage Reference VREF or Ve (With Increased

Line Impedance):
In order to show that the good conventional controller results of Figs. 4 and 5 depend
on operating conditions, the line impedance is increased to pu and, thereafter, the
previous 5% step change test is repeated. Increasing the line impedance represents the
case of one of two parallel transmission lines, or part of a ring connected power
system, being switched out. The results in Figs. 6 and 7 clearly show that the
conventional controller performance has degraded significantly compared to the
neurocontrollers, which give consistently good results even when conditions change.
In particular, the conventionally controlled rotor angle excursions in Fig. 7 are
considerably larger with less damping than in Fig. 5, because these linear controllers
were designed to have good damping characteristic for a system with different line
impedance.
Fig. 6. Terminal voltage for 5% step change in the Fig. 7. Rotor angle for 5% step change in the desired
desired terminal voltage with twice the transmission terminal voltage with twice the transmission line impedance
line impedance as in Fig. 3 (P = 1 pu and pf = 0.85 as in Fig. 4 (P = 1 pu and pf = 0.85 lagging).
lagging).
3) Short-Circuit Test:
In power systems, faults such as three-phase short circuits occur from
time to time, and because they prevent energy from the generator reaching the infinite
bus, it means that most of the turbine shaft power goes into accelerating the generator
during the fault. This represents a very severe transient test for the controller
performance. Figs. 8 and 9 show the terminal voltage and the rotor angle of a turbo
generator operating under the same conditions as in Figs. 4 and 5, and with the line
impedanceZ1, but with a temporary three-phase short circuit applied at the infinite
bus for 50 ms at t= 1 s. The system operating conditions prior to the fault once again
agree with those at which the linear conventional controllers were designed. The rotor
angle performance by the neurocontrollers in Fig. 9 is similar to that of the
conventional controllers, but in Fig. 8 the neurocontrollers give a significantly
improved terminal voltage response.

Fig. 8. Terminal voltage for a 50-ms three-phase short circuit

Fig. 9. Rotor angle for a 50-ms three-phase short circuit (P
(P = 1 pu and pf = 0.85 lagging). = 1 pu and pf = 0.85 lagging).
CONCLUSION:
This paper concludes that, the two separate COT neurocontrollers, one to
replace the AVR and the other to replace the governor, perform slightly better, but more
importantly allows flexibility in choosing the neurocontroller architecture learning rates.
In practice, this will translate into reduced computational demand. The neurocontrollers
consistently outperform the conventional linear AVR and governor, particularly when the
operating condition changes from that at which the linear controllers were designed. This
is to be expected since the power system is nonlinear and no stationary. The
neurocontrollers allow the turbo generator to either transmit more power over longer
transmission lines, and to withstand severe faults for longer durations than with the
conventional controllers. This could reduce the cost of upgrading existing lines or
increase the power per dollar invested. The successful performance of the COT
neurocontrollers, even when the system configuration changes, comes about because the
online training never stops, and deviation signals are used.

REFERENCES
 G. K. Venayagamoorthy and R. G. Harley, “A continually online trained ier for a
turbogenerator,” in Proc. IEEEhines and Drives Conf. (IEMDC’99), Seattle,WA,
May,9–12 1999, pp. 404–406.

 Q. H. Wu, B. W. Hogg, and G. W. Irwin, “A neural network regulator for turbo

generators,” IEEE Trans. Neural Networks, vol. 3, pp. 95–100, Jan. 1992.

 K. S. Narendra and K. Parthasarathy, “Identification and control of dynamical

systems using neural networks,” IEEE Trans. Neural Networks,Vol. 1, pp. 4–27,
Mar. 1990.
 Q. H. Wu, B. W. Hogg, and G. W. Irwin, “A neural network regulator for turbo
generators,” IEEE Trans. Neural Networks, vol. 3, pp. 95–100,Jan. 1992.

 Fundamentals of Artificial Neural Networks BY Mohammad H. Hassoun