Automatic Generation Control Using Artificial Neural Network

Based Narma L-2 Controller

Abstract
In this research two different types of controllers are implemented in two are hydrothermal
power stations to analyze the performance under step load disturbance. Controllers which are
used for this purpose are Artificial Neural Network (ANN) and PI controller. It is analyzed that
Artificial Neural Network gave better dynamic response and reduced error magnitude and
minimized frequency transient.
Introduction cause damage to the power generation and
An electric system consists of two main distribution system. The decrease in power
parts which includes generation and system frequency is assumed as avalanche in
distribution system. Subsystems are nature as it can cause outage for a long time
interconnected with help of tie lines. There due to damage in power stations.
is a matching requirement for each The outage due to avalanche nature can
subsystem for the generation and regulation cause loss in speed of exciter due to which
of system frequency which is known the EMF falls causing disconnection of
basic problem in load frequency control or consumers. Frequency avalanche cause
automatic generation control. AGC perform voltage drop which cause breakdown and
the role of division of load among different stoppage of power into paralleled station
systems which can help to gain maximum with interruption of power to the customers.
performance and economic advantages. The The function of automatic frequency control
mismatch in the generation and demand can to prevent the power system frequency from
cause frequency deviation which may lead approaching a critical value, when loss of
to system collapse. If the frequency of active power occurs, by disconnecting part
power system is less than standard it affects of the loads thereby keeping power stations
the quality of power. If the frequency of the and there auxiliaries operative .In this case
system is reduced by 0.5 to 2 Hz from he power system supplies to majority of
standard frequency, the steam turbines consumers suffer no interruption and system
undergo excessive vibration which can
to disconnect load can be restored within a The supplementary loop gives feedback via
fairly short period of time. the frequency deviation and adds it to the
Power System Frequency control
Frequency deviation is an immediate
aftereffect of the lopsidedness between the
electrical load and the dynamic power
provided by the associated generators. A
changeless off-typical frequency deviation
straightforwardly influences control
framework operation, security,
dependability, and productivity by damaging
turbines, degrading load performance, over-
primary control loop through a dynamic
burdening transmission lines, and setting off
controller. The resulting signal is used to
the assurance gadgets.
regulate the system frequency. In real-world
Since the frequency produced in the electric
power systems, the dynamic controller is
system is relative to the turn speed of the
usually a simple integral or proportional
generator, the issue of frequency control
integral (PI) controller. Following a change
might be specifically converted into a speed
in load, the feedback mechanism provides
control issue of the turbine generator unit.
an appropriate signal for the turbine to make
This is at first overcome by including an
generation (Δ Pm) track the load and restore
overseeing instrument that detects the
system frequency. The objective of an
machine speed, and modifies the
effective load shedding (emergency control)
information valve to change the mechanical
scheme is to curtail a minimum amount of
power yield to track the heap change and to
load, and provide a quick, smooth, and safe
reestablish frequency to a nominal value.
transition of the system from an emergency
There are three types of frequency deviation
situation to a normal equilibrium state.
controls in the power system which is shown
Performance of AGC under Normal &
in Figure 1.
Abnormal Conditions
1. Primary control
Under normal conditions with each area able
2. Supplementary control
to carry out its control obligations, steady
3. Emergency control
state corrective action of AGC is confined to system, so that u (t) = Kp e (t). Typically,
the area where the deficit or excess of given a step change of load demand, low
generation occurs. Inter area power transfers values of Kp give rise to stable responses
are maintained at scheduled levels and with large steady-state errors. Higher values
system frequency is held constant.[6] Four of Kp give better steady-state performance,
basic objectives of power system operation but worse transient response. Therefore, the
during normal operating conditions are higher value of Kp is used to reduce the
associated with automatic generation control steady state error, although increasing the
(AGC): gain Kp decreases the system time constant
and damping. [7]Therefore it is evident to
 Matching total system generation to choose the optimum value of Kp. The
total system load; proportional action can never eliminate the
 Regulating system electrical steady state error in the system because
frequency error to zero; some (small) error must be present in order
 Distributing system generation to produce a control output. A common way
among control areas so that net area of reducing the steady state error is by
tie flows match net area tie flow incorporating integral action into the
schedules; controller. Here, the control signal generated

 Distributing area generation among is proportional to the integral of the error

area generation sources so that area signal, that is,

operating costs are minimized. u ( t )=K i∫ e(t )dt

Proportional Integral Controller Here Ki is the integral gain. While an error
One of the most widely used control laws in exists, the integrator tends to increase
thermal and hydro power station governing control action, thus driving the plant output
systems is the PI type controller. towards the demand output. Then, when the
Proportional controller is used to reach the error disappears, the continuing integrator
steady state condition much quicker because output can be used to maintain the control
of the faster transient response with action necessary for steady-state conditions.
proportional controller. The proportional But, if the gain of integrator Ki is
term of the controller produces a control sufficiently high, overshoot will occur
signal proportional to the error in the increasing sharply as a function of the gain,
this is highly undesirable. Lower value of Ki Under steady state operation, the transferof
reduces overshoot but increases rise time of power over the Tie-line P12 can be written
the system. Based upon discussion it is as,
required to design both the Κp and Κi P 12=(│ V 1│ │V 2│/ X 12)sin( δ 1−δ 2)
properly Where V1 and V2 are the magnitudes of the
end voltages of control areas 1 and 2
respectively, and δ1 and δ2 are the voltage
and angles of V1 and V2, respectively
For a small change Δδ1 and Δδ2 in voltage
angles, the change in Tie-line power, ΔP12
is shown as
P 12=(│ V 1│ │V 2│/ X 12)cos (δ 1−δ 2)( Δδ 1−Δδ 2)
Synchronizing coefficient,
T 12=(│V 1 ││ V 2 │/ X 12)cos(δ 1−δ 2)
NARMA Controller
Various learning mechanisms exist to enable
the NN architecture that has been classified
into various types based on their learning
mechanisms and other features. Some
classes of NN refer to this learning process
as training and the ability to solve a problem
using the knowledge acquired as inference
Automatic Generation Control in Two
[9]. NNs have simplified limitations of the
Area System
central nervous system, and obviously
A power system comprising of two control
therefore, have been motivated by kind of a
areas interconnected by a weak lossless tie-
human brain termed as neurons are the
line is considered. Each control area is
entities, which perform computations such
represented by an equivalent generating unit
as cognition, logical inference, pattern
interconnected by a Tie-line with reactance
recognition and so on. Hence the
X12. Each area is represented by a voltage
technology, which has been built on a
source behind an equivalent source
reactance.
simplified limitation of computing by
neurons of brain, has been termed.
ANN controller architecture employed here
is Nonlinear Auto Regressive Model
reference Adaptive (NARMA). It is a
standard model that is used to represent
general discrete-time nonlinear systems. It is
simply a rearrangement of the neural
network plant model neural network plant
model is used to assist in the controller
training. This controller requires the least
computation. The only online computation is
a forward pass through the neural network
controller.
Artificial Neural Networks (ANNs) are
relatively crude electronic models based on
the neural structure of the brain. Artificial
neural networks try to mimic the functioning
of brain. In this study feed forward model is
used, which contains three layers; input, It consists of reference, plant output and

hidden and output layer. The ANN control signal. The plant output is forced to

controller takes two real valued inputs; ACE track the reference model output. The plant

and change in ACE, and gives a real valued model is used to predict future behavior of

output. Figure shows the structure of the the plant, and an optimization algorithm is

controller. The controller is designed with used to select the control input that

one input layer of two neuron, hidden layer optimizes future performance [11-12].

with three neurons and output layer with one For NARMA-L2 control, the controller is

neuron. simply a rearrangement of the plant model.

For model reference control, the controller is
a neural network that is trained to control a
plant so that it follows a reference model.
The drawback of this method is that the
plant must either be in companion form, or
be capable of approximation by a
companion form model. For training the
ANN Levenberg-Marquards back
propagation (TRAINLM) optimization
technique is used which is the fastest back
propagation algorithm and highly
recommended for supervised algorithm,
however, it requires more memory for
iterations. 1000 epochs are used for iteration
in this simulation. Activation function used
in the simulation is logsigmoidal.
Simulations and Results

A Plant identification window shows the no

of hidden layers training epochs as 100 are
used. Trainlem function is used for training
the network. Plant input output window
shows the outputs of the reference plant
model used for the NARMA controller to
train.

Conclusion
The settling time taken by NARMA
controller is less than PI controller. The
issue of overshoot caused by increasing the
gain of PI controller can also be removed by
using NARMA controller. The memory
required for saving data is less for NARMA
controller.
References
[1] Hadi Saadat, Power System Analysis, 2nd ed., McGraw-Hill, 2004, pp. 527-554.
[2] Olle I. Elgard, Electric Energy System Introduction,McGraw-Hill, New Delhi, 2005, pp.
219–361.
[3] A.R.Bergen, V. V. (2000). Power system analysis. PRINTICE HALL.
[4] I.J.Nagrth, D.P. Kothari (2010). power system analysis. Tata Mc Graw-Hill.
[5] Mohsen Rezaei Adaryani, Hossein Afrakhte,”NARMA-L2 Controller for Three-area Load
Frequency Control.” IEEE treansaction 2011.
[6] Neha Patel, P. B. (july 2013). Automatic Generation Control of three area power systems
using Ann Controllers. IJETAE ..
[7] K Chatterjee. Design of Dual Mode PI Controller for Load Frequency Control, International
Journal of Emerging Electric Power System. Vol. 11, Issue 4, Article 3.2010
[8] Elgerd O. I. 1982. Electric Energy System Theory: An Introduction. Mc-Graw Hill.
[9] C.Srinivasa Rao, “Design of Artificial Intelligent Controller for Automatic Generation
Control of Two Area Hydrothermal System,” International Journal of Electrical and Computer
Engineering (IJECE), vol. 2, No.2, pp. 183-190, April 2012
[10] Q. P. Ha. 1998. A Fuzzy sliding mode controller for Power System Load Frequency
Control. 2nd International Conference of Knowledge based Intelligent Electronic System.
Adelaide, Australia. 21-23 April.
[11] Kumpati S . Narendra, Kannan Parthasarathy, “identification and contol of dynamic systems
using neural network.” IEEE transaction on neural networks, vol-1, no-1, 1990.
[12] Rishabh Verma, Shalini Pal, Sathans, “ Intelligent automatic generation control of two area
hydrothermal power system using ANN and fuzzy logic.” IEEE, international conference on
communication system and network technologies, 2013.