1

1. INTRODUCTION
Load frequency is very important in power system operation and control for supplying sufficient and
reliable electric power with good quality. The generic functions of LFC include the following aspects
(i) Keeping frequency and tie line power constant and (2) considering Economic dispatch. Area load
changes and abnormal conditions such as outages of generation, lead to mismatches in frequency and
scheduled power interchanges between areas. LFC is an essential mechanism in electric power system
which balances generated power and demand in each control area in order to maintain the system
frequency at nominal value and the power exchange between areas at its scheduled value. Due to the
relatively fast loads demand fluctuations and the relatively slow area generation response rate, only
the instantaneous estimate of load demand cannot provide us with the satisfactory performance. A
look-ahead feature is needed if a more effective LFC is expected [15,16].
In this view, the key features of this paper are the development of a look ahead estimation for LFC
and Economic Load dispatch. Automatic generation control will act to adjust the output
generation every instant of time while economic load dispatch will adjust the participation
factors every few minutes to minimize the overall generating cost with respect to needed load
demand while keeping the frequency within the normal range. The idea is to estimate or
predict the load 10 to 15 minutes ahead of real time on intervals of 1 or 2 minutes. This
predicted load will be distributed the entire generator by minimizing the cost function of the
total load demand by adjusting the units participation factors of the generating units[1,14].
In this paper we have consider two control area in which first control area contains the
combination of hydro, thermal and gas generating unit and in second control area it contains
the combination of the thermal and hydro generating units. The short term load forecasting
using neural networks was proposed[16,21,22]. The predicted hourly load is being used to
provide the future load estimates to the LFC system. As load is constantly varying, the values
of gain of PI controller and the singular bounded hyper plane are required to change to get the
better response. But the amount of control required is easy for the operator to describe. This
required control can easily be described in qualitative terms and symbolic form. This
automatically leads the necessity of soft computing based intelligent controller like fuzzy
logic controller[6,7,12,13]. The fuzzy controller gives the opportunity to describe the control
action in qualitative terms. For this purpose in this paper Fuzzy Logic controller use for
LFC.
2. Modelling of the Interconnected Power System
For the system studies consider the combination of three and two generating units in first and
second control area respectively of thermal, hydro, gas and the combination of thermal and
hydro generating unit with their system regulation. Each control area is connected through
tie-line for their net balance interchanged tie-line power. The Transfer function model is
shown in Figure.1[3].

2


Figure 1. Block diagram of the interconnected power system.
3. Short Term Load Forecast using Neural Network
The total demand of a system is a equal to sum of all the consumers demand at the particular
time. The objective of Short term Load Forecasting(STLF) is to forecast the future system
load. A good understanding of the power system characteristics helps to design reasonable
forecasting models and select appropriate models operating in different situations. Because of
its importance, load forecasting has been extensively researched and a large number of
models were proposed during the past several decades, such as Box-Jenkins models, ARIMA
models, Kalman filtering models, and the spectral expansion techniques-based models[16-
28]. Generally, the models are based on statistical methods and work well under normal
conditions, however, they show some deficiency in the presence of an abrupt change in
environmental or sociological variables which are believed to affect load patterns. Also, the
employed techniques for those models use a large number of complex relationships, require a
long computational time, and may result in numerical instabilities. Therefore, some new
forecasting models were introduced recently. The development of Artificial Intelligence (AI),
Expert System (ES) and Artificial Neural Networks (ANN) have been applied to solve the
STLF problems[16,17,21,22,2325,28]. An ES forecasts the load according to rules extracted
from experts' knowledge and operators' experience. This method is promising, however, it is
important to note that the expert opinion may not always be consistent, and the reliability of
such opinion may be in question. Over the past two decades, ANNs have been receiving
considerable attention and a large number of papers on their application to solve power
system problems have appeared in the literature. This paper presents an extensive survey of
ANN-based STLF models. Back-Propagation multilayer feed-forward designing the NN
model. The typical BP network structure for STLF is a three-layer network with the nonlinear
sigmoid function as shown in Figure 2.

3








Figure 2: A typical BP network structure
.
The BP network is a kind of array which can realize non-linear mapping from the inputs to
the output variables. Therefore the selection of input variables of a load forecasting network
is of great importance. The input variables can be classified into 6 classes (i) Temperature (ii)
Hour of Day (iii) Day of the week (iv) Holiday (weekend indicator (0, 1) (v) previous 24 hr
average load and (vi) 168 hr (previous week) lagged load. The number of output variable is 1
i.e load at particular hour.

4. Five-minute ahead error predictor:

For improving the LFC performance a lookahead control algorithm is to be design. In this
purpose hourly estimates load from NN model can interpolate it to a 5 min sample rate. So
from this interpolation 24 hour forecast load data can be generated at 5 min interval. From
this data look-ahead error can be predicated. Suppose at time t the estimated load is y(t) and
after time (t+5) the estimated load is y(t+5). The error or load change within this 5 minute is
P
d
(t)=y(t+5)-y(t). If this load changes consider at time t, than 5 min look-ahead error
incorporate in the LFC system and corresponding design the control algorithm. This total
load change divided all the units by concept of the units participation factor.

5. Generation Allocation of the Generating Unit
Every control area of the interconnected power system may contains single or many
generating units of different driving source(hydro, thermal, nuclear, gas etc) which insure to
provide the stable frequency and net balance interchanged tie-line power flow between the
different control area via coordination with the calculation of automatic generation control or
control mechanism and economic load dispatch. This coordination insuring to provide the
information and logic that how much each generating unit will participate or take the load
sharing out of the calculation of total load demand in context with their economics. This logic
lead to introduce the concept of the units participation factor. Participation factor defined as
the rate of change of each units output with respect to a change in total generation of the that
control area. From the definition and properties of the participation factor its summation is
X
x
1
x
2



X
6

Hidden
Layer
Input
Layer
Output
Layer
4

equal to unity for each respective control area. . When the economic load dispatch calculation
is performed then the sum of the present unit generation equal to the total generation then it
will assigned as the base point generation. This base point generation will be the most
economic output of the every generating unit. Thus we will assigned the generation allocation
of the multiple generating unit with respect to total change in generation output with the help
of the conjunction of the load frequency control mechanism and the concepts of economic
load dispatch.

(1)

desired output from unit i

change in total generation

= base-point generation for unit i

= participation factors for unit i

(2)
To find out the participation factors of each unit, we use the concepts of economic load
dispatch calculation and the following equations accomplish the execution task. The
quadratic cost function of the power system is given as

(3)

and

are the coefficient of the cost function and

is power generation of the

unit.
()

) (4)
Power balance constraint is given as

(5)
Where,

load demand and

= transmission losses and inequality is given as

(6)

are their minimum and maximum power generation of the

unit.
5


Figure 3. Generation allocation block diagram of the generating unit


6. Application of PI like Fuzzy controller for Load Frequency Control

The selection of the control variables (controller input and controller outputs) depends on the
nature of the controlled system and the desired output. In this work we have designed PI- like
Fuzzy Knowledge Based Controller. The basic structure of the conventional PI controller is
}
+ = e e u
I P
k k dt (7)
where
p
and
I
are the proportional and integral gains respectively and e is the error
signal(i.e. e=process set point process output variable). Taking the derivative with respect to
time, the above expression 7 is transformed into equivalent expression
.
u =
p
.
e +
I
e (8)
For the load frequency control problem the inputs to the Fuzzy controller for i
th
area at a
particular instant t are ACE
i
(t) and ACE
i
(t) , where ACE
i
(t)=P
tiei
+B
i
f
i
and ACE
i
(t) =
ACE
i
(t)- ACE
i
(t-1) and output of the fuzzy controller is u .This is in accordance with the
eqn. (8) for PI like controller. The inputs and output are transformed to seven linguistic
variables NB, NM, NS, Z, PS, PM and PB which stand for Negative Big , Negative Medium,
Negative small, Zero, Positive Small, Positive Medium and Positive Big respectively .
Symmetrical triangular (expect of the two outermost ones which have a trapezoidal shape)
membership function is considered here for all the three variables of ACE, ACE and u.
The membership function of ACE over the operating range of minimum and maximum value
of ACE is shown in Figure 4. The membership function would perform a mapping from the
crisp value to a fuzzified value. As shown in Figure 4, one particular crisp input ACE is
converted to fuzzified value i.e.
NM NS
2 . 8 .
+ ,where 0.8 and 0.2 are membership grades
6

corresponding to the linguistic variable NS and NM in FNN system. The membership grades
are zero for all other linguistic values except NS and NM. The crisp value input to the system
in this way is converted to a fuzzified value consisting of several membership grades
corresponding to each linguistic variable. In the same manner, the other input, ACE, and the
output u, are fuzzified.

Figure 4. Membership functions for the fuzzy variables of ACE
A. Procedure for Framing the Rules
As each of the three fuzzy variables are quantized to seven fuzzy sets, so total 7 7 or 49
rules are required to generate an fuzzy output relating two input fuzzy sets as shown in Table
1. Fuzzy rules play a major role in FLCs and have been investigated extensively.

TABLE-1: FUZZY RULES FOR TWO-AREA SYSTEM

However, fuzzy rules usually can be generated using knowledge and operation experience
with the system or by understanding of the system dynamics [7]. Table-1 shows the 49 rules
that are generated through the behavior of the system response after a load change. The
entire rule base may be divided into regions A, B, C and D depending on the dynamics of
the system as shown in Figure 4. From Figure 5, if e is positive and e is negative, the system
will reduce the error itself, which means that the magnitude of the applied control action
should be zero. This corresponds to the region labeled A in the Figure 5. In case of negative e
and negative e the system tends to go to instability which requires an opposite (positive
control) control action. This is interpreted as region B in the figures. In region C, the error is
still negative while the rate of change of error is positive. This implies that the error is
decreasing and that the control action should be kept to a minimum (zero).


7


Figure 5: Rules Generation by Understanding the System Dynamics

When both the error and rate of change of error go positive, the system again tends to go to
instability region again and hence it is required to apply opposite (negative) control action to
compensate for this tendency toward instability.
The output control action increases from zero (Z) control in rule region A to a
minimum positive control (PB) in rule region B, as indicated in Figure 5. Then the control
action decreases from PB in rule region B to minimum control (Z) in rule region C. Also
when both e and e approach their maximum positive value, the control action goes to
maximum negative (NB).
The firing strength of a fuzzy control rule will be ultimately determined using
Mamdani correlation minimum encoding inference. For example, suppose that at a certain
instant t, the Fuzzy logic employs a rule:
If ACE (t) is NS and ACE (t) is NM then control action u is PS.
The firing strength of such fuzzy control rule will be,
B A
(ACE(t)
,
ACE(t)) =

min(
ACE
(ACE(t),
ACE
(ACE(t)), where, A: NS,B: NM,
ACE
and
ACE
represent the membership functions of the linguistic values of error and
rate of change of error. Suppose, e belongs to NS with a membership of 0.3 and e belongs
to NM with a membership of 0.7, then the rule consequence (w
i
) will be 0.3, the minimum
between 0.3 and 0.7.
B. Defuzzification
The output of the inference mechanism is a fuzzy value, so it is necessary to convert this
fuzzy value into a real value, since the physical process cannot deal with fuzzy value. This
operation which is the inverse of fuzzification is known as defuzzification. The well-known
center of gravity defuzzification method has been used because of its simplicity. The control
output u A is determined using the center of gravity by the following expression,

8

=
=
= A
49
1
49
1
j
j
j
j j
u
u

(9)

Where,
j
is the membership value of the linguistic variable recommending the fuzzy
controller action, and
j
u is the precise numerical value corresponding to that fuzzy controller
action. This Au is added with the existing previous signal will be the actually output signal u
which goes to the governor. Figure 2 is the block diagram of two area system where area
control error of each area is fed to the corresponding fuzzy logic based digital controller. The
accurate control signal is generated for every incoming ACE.
II. NORMALIZATION
Generally the universes of discourse of input and output variables of FLC are the real line. In
practice, each universe is restricted to an interval that is related to the maximal and minimal
possible values of the respective variable, that is, to the operating range of the variable. The
rules will not be properly framed if the operating range is not suitably selected. For
simplification and unification of the design of the FLC and its computer implementation,
however it is more convenient to operate with normalized universe of discourse of the input
/output variables of the FLC [19]. Figure 6 shows block diagram representation of Fuzzy
logic Controller with normalization and denormalization factors. The normalized universes
are well-defined domains; the fuzzy values of the input/output variables are fuzzy subsets of
these domains. In general, the normalized universe can be identical to the real operating
ranges of the variable, but in most application they coincide with the closed interval of (-
1,1). In this work, the inputs of ACE and ACE (t) are normalized between are -1 to 1 the
real operating universe to normalized universe. Scaling factor is dependent on the operating
range considered for the corresponding input. As for example, K
ACE
=
02 . 0
1
=50, as the
operating range of ACE is considered as -0.02 to 0.02. Essentially, the scaling factors K
ACE

and K
ACE
map the real measured values ACE(t), ACE(t) to the values of ACE
*
(t),
ACE
*
(t) from these normalized universes by linear mapping such as, ACE
*
= K
ACE
ACE(t);
ACE
*
(t) = K
ACE
ACE(t). Defuzzified value u
*
(t) obtained by the application of the FLC
algorithm, belongs to the normalized universe (-1, 1) and it is required to get the real change
of control variable u (t) from these operating range (-1,1) through the scaling factor K
u
.
Therefore the output u(k) = K
u
u
*
(t). Thus, the scaling of the normalization value of the
output variable is effectively denormalization, that is, bringing the output of the FLC from the
normalized universe to its actual operating range (-0.01, 0.01) through the scaling factor K
u
=
1
01 .
= 0.01.

9


Figure 6: Fuzzy Logic Controller with Normalization and Denormalization factors

7. Results and Discussion
The proposed design dual mode controller in conjunction with economic load dispatch (i.e.
units participation factor) has been applied for typically multi generating units of the
interconnected power system..The proposed controller is simulated in MATLAB
SIMULATION TOOLBOX. A neural network based very short term load forecasting system
is developed in this paper .The models are trained on hourly data and 35064 data sets is used
in each epoch. The training parameters were set with initial learning rate, momentum
constant and error convergence0.1, 0.01 and 10
-7
respectively. Total no. of epochs for training
is 500. 1754 number of sample data sets use for testing the network. The models are shown
to produce less average error during the forecasting the load. The forecasted load pattern is
dashed line and actual load pattern in solid line shown in Figure 7. From this forecasted
loadcurve a 5 minute look-ahead load change is generated. The error or load change curve
shown in the Figure 8. This load change now fed to the area 1. Each generating unit now
participate in the load change by using the participation factor of each unit.The dynamic
response or frequency deviations of area 1 and area 2 are shown in the Figure 9 and 10
respectively. The proposed control algorithm or switching scheme (SS) works by switching
between proportional controller and integral controller depending on the rate of change of the
error signal. The proposed controller gives better dynamic response respect to consider a
fixed gain integral controller.
10


Figure 7: Actual and Forecast Load Pattern

Figure 8: Load Change (Error) signal fed to area 1
0 5 10 15 20 25
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
x 10
4
Hour
L
o
a
d


actual Load
Forcast Load
0 200 400 600 800 1000 1200 1400
-0.1
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
time(min)
L
o
a
d

C
h
a
n
g
e
(
M
w
)
11


Figure 9: Frequency deviations in area-1

Figure 10: Frequency deviations in area-2
0 500 1000 1500
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
time(min)
f
r
e
q
u
e
n
c
y

d
e
v
i
a
t
i
o
n
s
(
H
z
)


a
r
e
a
-
1
0 500 1000 1500
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
time(min)
f
r
e
q
e
n
c
y

d
e
v
i
a
t
i
o
n
s
(
H
z
)


a
r
e
a
-
2
12

8. Conclusion
This paper describes a methodology for estimating the look-ahead error in 5 min-steps for a
two-area interconnected power system. Each units share the load according to their
participation factors with the total load change of the control area. This paper builds a fuzzy
rule base with the use of the area control error and rate of change of the error. The simulation
results for continuous load change have been justified that the proposed Fuzzy Logic
Controller yields more improved control performance than a fixed gain integral controller.
Nomenclature and system parameters
Steam Turbine:

(Coefficient of re-heat steam turbine) = 0.3,

(re-heat time constant) = 10 sec.

(Turbine time constant) = 0.3 sec,

(speed governor regulation) = 2.4 Hz/pu MW

(Speed governor time constant) = 0.08 sec.,

(wate time constant) = 1.0 sec

(Speed governor regulation) = 2.4 Hz/pu MW

Gas Turbine:
(Speed governor lead time constant) = 0.6 sec, ((speed governor lag time constant) = 1.0
sec., Valve positional constant = 1, = 0.05 and = 1,

(Fuel time constant) = 0.23 sec.

(Combustion reaction time delay) = 0.3 sec.,

(Compressor discharge volume time

constant) = 0.2 sec.,

(Speed governor regulation) = 2.4 Hz/pu MW

Hydro Turbine:

(Speed governor rest time) = 5.0 sec.,

(Transient droop time constant) = 28.75

(Main servo time constant) = 0.2 sec.

References:
[1] A.J.Wood and Woolenberg .B.F. Power Generation Operation & Control,John
Wiley& sons, 1998.
[2] O.E. Elgerd, Electric Energy Systems Theory(second Edition),Newwork
:McGrawHill, Inc.1982,pp 315-389.
[3] K. S. S. Ramakrishna, Pawan Sharma, T. S. Bhatti, Automatic generation control of
interconnected power system with diverse sources of power generation, International
Journal of Engineering, Science and Technology, vol.2, no.5, 2010,pp.51-65
[4] M.L.Kothari, J.Nanda, D.P.Kothari and D.Das. Discrete-mode Automatic generation
control of a two-area reheat thermal system with new area control error, IEEE
Transactions on power systems,Vol.4, No.2,May 1989,pp.1988-1994.
[5] S.C.Tripathy, G.S.Hope and O.P.Malik. Optimization of load-frequency control
parameters for power systems with reheat steam turbines and governor dead band
nonlinearity, IEE Proceedings, Vol. 129,Pt.C,No.1, Jan 1982,pp.10-16.
13

[6] G.A. Chown and R.C.Hartman. Design and experience with a fuzzy logic controller
for Automatic generation control, IEEE Transactions on power systems Vol. 13 ,No.
3, August 1998,pp.965-970.
[7] Jawad Talaq and Fadel Al-Basri. Adaptive fuzzy gain scheduling for load frequency
control, IEEE Transactions on power systems, Vol. 14, February 1999,pp.145-150.
[8] T.P.Imthias Ahamad,P.S. Nagendra Rao and P.S.Sastry. A reinforcement learning
approach to automatic generation control, Electric power systems research 2002(63),
pp.9-26.
[9] Y.L.Karnavas and D.P. Papadopoulos. AGC for autonomous power system using
combined intelligent techniques, Electric power systems research 2002(62),pp.225-
239.
[10] Kundur. P, Power System Stability and Control, McGraw Hill, New York, 1994.
[11] Dynamic Models for steam and Hydro Turbines in Power system studies,IEEE
committee report. Transactions in Power Apparatus & Systems
Vol.92,No.6,Nov./Dec.1973,pp.1904-915.
[12] Mohamed E.El-Hawary, Electric Power Applications Of Fuzzy Systems,IEEE Press
series on Power Engineering chapter 5 pp.112-148.
[13] Roland R.Yager and Dimitar P.Filev,Essentials of Fuzzy Modelling and Control,John
Wiley & Sons,Inc.1994H S. Farook, P. Sangamasmeswara Raju optimization of
feedback controller power system using Evolutionary Genetic Algorithm, International
Journal of Engg. Science & Tech; vol.3 , No.5 May 2011
[14] O. I. Elgerd and C. Fosha, optimum megawatt frequency control of multi-area electric
energy systems, IEEE Transactions on Power Apparatus and Systems, vol. PAS-89, no.
4, pp. 556-563, Apr. 1970.
[15] Danial J. Trudnowski, Warren L. McReynold, and Jeffery M. Jonson, Real-Time Very
Short-Term Load Prediction For Power-System Automatic Generation Control, IEEE
Trans. On Control Systems Technology, Vol. 9, No. 2, March 2001.
[16] K. Lui, S. Subbarayan, R.R. Shoults, M.T. Manery, C. Kwan, F.L. Lewis, J.
Naccarion, Comparison of Very Short-Term Load Forecasting Techniques, IEEE
Trans. on Power System, Vol. 11, No. 2, May 1996.
[17] Ibrahim Moghram, Saifur Rahman, Analysis and Evaluation of Five ShortTerm Load
Forecasting Techniques, IEEE Trans. on Power System, VOl. 4, No. 4, Oct. 1989.
[18] J. P. Rothe, Dr. A. K. Wadhwani,. S. Wadhwani, Short Term Load Forecating Using
Multi Parameter Regression, International Journal of Computer Science and
Information Security, Vol. 6, No. 2, 2009.
[19] H.Danishi, Member, IEEE, and A. Daneshi, Real Time Load Forecasting in Power
System, DPRT 2008, 6-9 April 2008, Nanjing China.
[20] R.P. Shulte, An Automatic Generation Control Modification For Present Demands on
Interconnected Power Systems, IEEE Trans. on Power System, Vol. 11, pp. 1286-
1294, 1996.
[21] R.C. Bakirtzis, V. Petridis, S.J. Klartiz and M.C. Alexladls, A Neural Network Short
Term Load Forecasting Model for Greek Power System, IEEE Trans. Power Syst.,
Vol 11, pp. 858-863, May 1996.
[22] H. S. Hippert, C.E. Pedriera, and R.C. Souza, Neural Networks for Short-Term Load
Forecasting: A Review and Evaluation, IEEE Trans. on Power System, Vol. 16, No. 1,
pp. 44-55, 2001.
14

[23] C. N. Lu, and S Vemuri, Neural Network For Short-Term Load Forecasting, IEEE
Trans. on Power System, Vol. 8, No. 8, pp. 336-342, 1993.
[24] T. Senjyu, Uazato. T, Higa P, Future Load Curve Shaping Based on Similarities Using
Fuzzy Logic Approach, IEE Proc. of Generation, Transmission, Distribution, Vol. 145,
No. 4, pp. 375-380,1998.
[25] A. G. akirtzis, J. B. Theocharis, S. J. Kiartzis, and K. J. Satsios, Short Term Load
Forecasting using Fuzzy Neural Network, IEEE Trans. on Power System, Vol. 10, No.
3 , pp.1518-1524, 1995.
[26] Muhammad Riaz Khan, Ajith Abraham, Short Term Load Forecasting Models in Czech
Republic Using Soft Computing Paradigm, International Journal Of Knowledge-Based
Intelligent Engineering system, IOS Press Netherlands, Vol. 7, No. 4, pp. 172-
179,2003.
[27] Khaled M. EL-Naggar, and Khaled A. AL-Rumaih, Electric Load Forecasting Using
Genetic Based Algorithm, Optimal Filter Estimator and Least Error Squares technique:
Comparative Study, World Academy of Science, engineering and Technology 6, 2005
[28] A. Asar and J. R. Mcdonald , A Specification of Neural Network Application in the
Load Forecasting Problem , IEEE Trans. on Control System technology, Vol. 2, No. 2,
1994, pp. 135-141.
[29] Mohamed E.El-Hawary, Electric Power Applications Of Fuzzy Systems,IEEE Press
series on Power Engineering chapter 5 pp.112-148.
[30] Roland R.Yager and Dimitar P.Filev,Essentials of Fuzzy Modelling and Control,John
Wiley & Sons,Inc.1994.