Beruflich Dokumente
Kultur Dokumente
(5)
(6)
(1)
Where σ is the coefficient of dispersion and is given
by:
(2)
193
3.
The speed control of the IM by an If e is Ai, and Δe is Bi, then ' k p is Ci, and ' ki is Di.(12)
adaptive controller FLC-pi Where Ai, Bi ,Ci and Di are fuzzy sets on
corresponding supporting sets.
In this paper PID parameters controller can be The fuzzy rules in (12) may be extracted from
operator’s expertise or based on the step response of the
adjusted by an adaptive mechanism based on a fuzzy
process.. By using the membership functions, we have
inference (adaptive FLC-PI). the following conditions
194
GA’s are parallel and global search techniques basic idea is to maintain a population of chromosomes
which take the concepts from evolution theory and (representing candidate solutions to
natural genetics to evolve solutions to problems. The
the concrete problem being solved) that evolves over
time through a process of competition and controlled 4.2.3. Optimization algorithm using GA of the
variation. GA’s is theoretically and empirically Fuzzy adapter
proven to provide robust search in complex spaces,
giving a valid approach to problem requiring efficient The application of the GA in the optimization
and effective searching. process of the FL controllers can be formulated as
• Evaluation of individual fitness; follows:
• Formation of gene pool (intermediate population); 1. Start with an initial population of solutions that
• Recombination and mutation. constitutes the first generation (P (0)).
2. Evaluate P (0):
4.2. Design of fuzzy-genetic system a) Take each chromosome (KB) from the population and
introduce it into the FLC,
4.2.1. Membership parameters optimization b) Apply the FLC to the controlled system for an
adequate evaluation period,
GA is applied to modify the membership c) Evaluate the behavior of the controlled system by
functions. When modifying the membership producing a performance index to the KB.
functions, these functions are parameterized with one 3. While the termination condition is not met, do
to four coefficients (Fig. 3), and each of these
coefficients will constitute a gene of the chromosome
for the GA.
c) t = t+1.
4. End.
The mechanism of this optimization procedure can be
represented in fig. 4.
input
(trial-and error) and one input (command action)
variables. A chromosome is formed from the decision Fig. 4 Evolutionary learning of an FLC
table by going row-wise and coding each output
fuzzy set as an integer in 0, 1… n, where n is the The fuzzy adapter consists of two inputs (error and its
number of membership functions defined for the derivative) and two outputs ( k p’ ) and ( ki’ ), where each
output variable of the FLC. Value 0 indicates that input has seven membership functions. These subsets are
there is no output, and value k indicates that the labeled by linguistic terms such as: Zero (Z), Negative
output fuzzy set has the k-th membership.
195
(N)... etc. GA is used to search the appropriate Fig. 5 Membership Fig. 6 Membership functions
parameters values and to modify the decisions table of functions e of Δe
of the FLC, where the chromosome is formed from
the decision table and to code each membership The resulting rule bases from the optimization procedure
function by a integer number from 0 to 2, number 2 are shown in table I and II. In the tables for example, the
indicates the number of membership function defined first rule for the output kp’ and k is:
for the two outputs. So, we can present the equivalent If e is A11 And e Δ is A21 So kp’ is B12 and ki’ is B21
code by: Small (S): 1, Big (B): 2 and No output: 0. In Where B12 is the second fuzzy set of the first
GA, we only need to select some suitable parameters, Consequent ( k p’ ) and B21 is the first fuzzy set of the
such as generations, population size, crossover rate, second consequent ( ki’ ). We can re-write this rule as:
mutation rate, and coding length of chromosome, If e is NB And e Δ is NB So kp’ B and ki’
then the searching algorithm will search out a is S.
parameter set to satisfy the designer's specification or
the system requirement. In this paper, GA will be
included in the design of fuzzy gains tuner of the PI
controller. The parameters for the GA simulation are
set as follows: The parameters for the GA simulation
are set as follows:
(1) Initial population size: 30;
(2) Maximum number of generation: 100;
(3) Crossover: Uniform crossover with probability
0.8;
(4) Mutation probability: 0.01.
In this paper, the performance is measured using the
following criteria.
(5) Minimum integral of squared which is given as
follows:
196
Fig. 10 Simulated results comparison of adaptive PI using
fuzzy inference and adaptive PI using fuzzy inference
optimized by GA of IM
5. Simulation results
6. Conclusion
To prove the rightness and effectiveness of In this work, we proposed a method of
proposed control scheme, we apply the designed combination between the fuzzy controller and
controller to the control of the induction motor. It conventional PI controller in order to overcome the
mainly consists of an induction motor, a ramp disadvantages of PI controllers and FLC, this combination
gave us an adaptive PI controller which presented
comparison current-controlled pulse width modulated
satisfactory performances (no overshoot, minimal rise
(PWM) inverter, a slip angular speed estimator, an time, best disturbance rejection). The major drawback of
inverse park, an outer speed feedback control loop and the fuzzy controller is the insufficient analytical design
a fuzzy gain scheduling of PI controller or fuzzy gain technique (choice of the rules, the membership functions
scheduling of PI controller optimized by GA for the and the scaling factors). That we chose with the use of the
speed control. Fig.8 to Fig.10 show the various genetic algorithm for the optimization of this controller in
order to control IM speed. In the system, GA is used to
simulation results.
design an adaptive PI controller using fuzzy controller
with optimal parameters. The optimal fuzzy gains
scheduling of PI controller is used to achieve robust
performance against parameter variations and external
disturbances. Simulation results have shown that the
proposed optimal controller is robust with regard to
parameter variations and external load disturbance (no
overshoot, minimal rise time, best disturbance rejection).
Finally, the proposed controller provides drive robustness
improvement and assures global stability.
Fig. 8 The IM rotor speed control with adaptive FLC- Appendix:
PI optimized by GA for two different Rr.
References:
197
2. FAULT DETECTION FOR POWER TRANSFORMERS
USING NEURAL NETWORKS
NAGIREDDY RAVI1 N. Yadaiah2MIEEE
1
Mini Hydel Stations, Pochampad, Andhra Pradesh Power Generation Corporation Limited, POCHAMPAD– 500
082, A. P., INDIA, E-mail: nagireddyravi@rediffmail.com
2
Dept. of Electrical and Electronics Engineering, College of Engineering ( Autonomous), Jawaharlal Nehru
Technological University, ANANTAPUR- 515 002, A. P., INDIA, E-mail: yadaiahn@hotmail.com
Abstract- This paper presents the methodology for The faults that occur within transformer
incipient fault detection in Power transformers using broadly classified into two types, one is an internal
artificial neural networks. An artificial neural incipient faults and other is an internal short circuit
network is used for identifying the practical
faults. The majority of incipient faults occurring in
transformer faults. The conventional Dissolved Gas
Analysis method to detect incipient faults has been power transformer give evidence very early in their
improved using artificial neural networks and is development stages through the transformer oil gas
compared with Rogers ratio method with available analysis. The transformer may function well
samples of field information. externally, it may cause serious problem with some
Key words: Incipient faults, Power Transformers, insulation deterioration. The fault detection
Neural Networks. techniques such as Dissolved Gas Analysis (DGA)
[1-2] and partial discharge analysis [3] are currently
used for detection of incipient faults. In this paper, a
I. INTRODUCTION method based on DGA has been improved using
The transformer is an important component Artificial Neural Networks (ANN) to identify the
in power system network. The failures of the faults. It is an off-line technique and requires DGA
transformer hinder the proper operation of electrical report. In addition to this, it is high in cost and time
system network and its reliable operation is an
important factor. Any fault in the transformer side consuming. The performance of the ANN based
causes power outages and blackouts, thereby hampers method has been compared with the existing method
the power quality. Power quality is one of the main such as Rogers ratio by using the field information of
concerns of the utilities and inferior power quality DGA samples from Andhra Pradesh (India)
causes many problems such as malfunctions, Distribution, Transmission, Generation Companies.
instabilities, short lifetime and so on, hence to
maintain power quality is most important. The
replacement of power transformer is very expensive
and time consuming and therefore it is essential to II. PROBLEM FORMULATION
detect incipient faults as early as possible, which
enable rectification of fault with minimum In this paper, the problem of identification
interruption of service. of internal incipient faults in power transformer has
been considered. Currently, DGA based methods are
198
being used, but these cannot provide accurate results. described below have proposed for identification of
The uncertainty in the process of sampling which is incipient faults in the power transformers.
due to improper collection of oil samples, transport of
An ANN method based DGA has been used
oil samples to laboratory. The errors in measuring
to overcome the limitations of existing methods. To
and analyzing in the laboratory due to leaks in the
implement the proposed method, the DGA data of oil
vacuum system during gas extraction and
samples for different transformers in Andhra Pradesh
contamination of columns used in gas
(India) Generation, Transmission and Distribution
chromatograph. The involvement of group discussion
Companies, which contains no fault and faulty
by experts is necessary to achieve a conclusion and to
samples, has been collected. This data has been used
make decision for disassembly. Due to this time
to train the ANN and the performance of this method
consuming and tedious procedure, automatic methods
has been compared with conventional method.
III. ARTIFICIAL NEURAL NETWORKS shown in Table I. The ratios of these gases are used
for interpretation of faults.
Artficial neural networks composed of
simple elements operating in parallel [7]. These
elements are inspired by biological nervous system.
An artififial neuron with activation function is shown
in Fig.1.
TABLE I
DISSOLVED GASES IN OIL.
199
dissolved gases of transformer oil. The limitation of Characteristic Fault Type Code of
this method is that, it can detect only three faults and characteristic
is very complex. gases
No Fault 0 0 0
Low temperature fault (<1500C) 0 0 1
Low temperature fault (1500C- 0 2 0
B. Rogers Ratios Method 3000C)
This method has been proposed by Rogers Medium temperature fault 0 2 1
and utilizes gas ratios listed in Table III and it reveals (3000C-7000C)
five different types of faults such as (i) Low High temperature fault (>7000C) 0 2 2
temperature thermal fault, (ii) Medium temperature Low energy partial discharges 0 1 0
thermal fault (iii) High temperature thermal fault, (iv) High energy partial discharges 1 1 0
Partial discharges and (v) High Energy arcing. It is Low energy discharges 1-2 0 1-2
one method most commonly used for diagnosis of the High energy discharges 1 0 2
transformer incipient faults.
TABLE II C. ANN based fault detection
GAS RATIOS FOR DORENENBERG RATIOS An Artificial Neural Network is good tool
METHOD for the applications like pattern classification, fault
Ratio1 (R1) CH4/H2 detection; function approximation due to inherent
Ratio2 (R2) C2H2/C2H4 characteristics of learning, generalization, fault
Ratio3 (R3) C2H2/CH4 tolerance etc. In this paper it has been used for off-
Ratio4 (R4) C2H6/ C2H2 line fault detection of power transformers incipient
fault with knowledge of DGA of transformer oil
samples. The problem of fault detection is a potential
application of ANN [5-6]. A multi-layer feed-forward
TABLE III
artificial neural network has been proposed to
GAS RATIOS FOR ROGERS RATIOS
identify the fault and a generalized delta rule has
METHOD
been used for training of network. Implementation
involves the generating of training data, and design of
Ratio1 (R1) CH4/H2 ANN structure, training of network and testing. In
Ratio2 (R2) C2H2/C2H4 this application, gas ratios become the inputs for
Ratio3 (R3) C2H4/ C2H6 neural network and types of faults become the
outputs of the network. To start with, a three layer
network has been considered, in which size of input
The code generation of each Rogers ratios
layer is equal to number of gas ratios and size of
for classification is described in Table IV and it was
output layer is equal to the number of different faults
defined by Rogers [4]. The coding of each fault type
and the nodes in the hidden layer are selected based
with respect to their characteristics is given in Table
on experience and a schematic diagram of a three
V.
layer network is shown in Fig. 2.
TABLE IV
CLASSIFICATION TABLE BY ROGERS
C2H2/ C2H4 CH4/H2 C2H4/ Range of
C2H6 Gas
Ratios
0 1 0 <0.1
1 0 0 0.1-1.0
1 2 1 1.0-3.0
Fig .2 Multi-layer feed-forward Neural Network.
2 2 2 >3.0
D. Algorithm
TABLE V : TYPES OF FAULTS Step 1: The data of oil samples of dissolved gas
analysis of various normal and faulty transformers
200
are collected. The data collected is used for training 4 Low partial energy 13
the neural network. discharge (LPD)
Step 2: The faulty type samples have been confirmed
by actual faults. The ANNs were trained with 472 training
Step 3: The ANNs are established on these samples samples. The three gas ratios are considered as inputs
using multi-layer back-propagation. and five nodes in output layer to identify five
Step 4: The inputs are the gas ratios of the DGA test different types of faults, in which the nodes are set to
report , the outputs are the fault types for the neural 1 for respective fault and 0 for no fault. After
network. training, the network is tested with typical test
Step 5: The trainining of the network is carried out samples and the results are summarized in Table VII.
with all training pairs from the training set by The performance of both the methods is compared
applying the input vector to the network input. and the results show that the ANN method detects
Step 6: Calucate the output of the network more accurately than Rogers ratio, method as shown
Step 7: Calculate the error between the network in Table VII.
output and the desired output i.e, the target vector Table VII : MPARISON OF ANN METHOD
from the traing set. WITH ROGER’S RATIO METHOD
Step 8: Adjust the weights of the network for Sa C C2 C H2 C2 A Rog Act
minimizing the error. m 2 H4 H H6 N er’s ual
Step 9: Repeat steps 1 through 4 for each vector in ple H 4 N Rati Fau
the training set until the error for the entire set is no 2 Me o lt
acceptably low. tho met
Step 10: The trained network is tested using test data. d hod
D. Simulation Results 1 1. 12 25 27 16 NF NF NF
0 5 .0 .2 9.1
In order to illustrate these methods such as Rogers 9 8 1
ratios method and ANN based method, experimental 2 0 0. 4. 21 11 NF NF NF
results of DGA for various power transformers have 98 86 .5 3
been collected from the Andhra Pradesh Power 4
Distribution, Transmission, and Generation 3 1. 0. 3. 11 12 NF UNI NF
companies. The input vector consists of three gas 9 67 12 .8 0.8 DE
ratios such as (i) C2H2/C2H4, (ii) CH4/H2, and (iii) 8 2 NTI
C2H4/C2H6 and the output vector involves five FIA
elements (i) No fault, (ii) Low Temperature fault BLE
(1500C - 3000C), (iii) Medium Temperature fault 4 0 14 5. 69 23. NF UNI NF
(3000C – 7000C), (iv) High temperature fault 5. 92 .0 09 DE
(>7000C) and (v) Low energy partial discharge. The 6 5 NTI
collected samples are listed in Table VI with FIA
respective fault type and these samples are used to BLE
train ANN. The fault type samples were confirmed 5 0. 2. 74 24 74 LT LTF LT
by the actual faults in the internal examination of the 2 67 .5 .2 F F
transformer. 3 9 8
TABLE VI : COMPOSITION OF THE TRAINING 6 3. 8. 76 34 52. LT UNI LT
DATA 8 74 .9 .8 89 F DE F
6 3 8 NTI
Fault Fault Type Training
FIA
Code Samples
BLE
0 No fault (NF) 400
7 1 16 67 11 96. M MT MT
1 Low temperature fault 25
6 .2 .6 23 TF F F
(150oC-300oC) (LTF)
5
2 Medium temperature fault 14
8 1. 45 90 54 2.1 HT HTF HT
(300 0C-7000C) (MTF)
8 6. .3 .1 9 F F
3 High temperature fault 20 2 2 7
(>7000C) (HTF)
201
9 0. 60 39 45 7.9 HT UNI HT Relations", Electronic Modeling, Vol. 24, No. 3,
3 .3 .3 .4 2 F DE F 2002, pp 48-61.
4 2 1 2 NTI
[7] N. Yadaiah, L.Sivakumar, Nagireddy Ravi
FIA
BLE "Design of Load Frequency Control in Power
10 3. 15 9 29 21 LP LPD LP Systems Using Neural Networks", Electric
9 4. 7 1.6 D D Energy Systems Management Proc. – Indian
7 5 7 Scenario, Roorkee, Sep 4-5, 1998, pp 134 - 138
V. CONCLUSIONS [8] MATLAB Documentation – Wavelet Tool Box,
An ANN method has been developed for Ver. 7, The Mathworks Inc., Natick, MA.
fault detection of power transformers. The
performance of ANN method has been tested and
compared with Rogers ratios method and found that,
ANN detects more accurately.
ACKNOWLEDGMENT
N. Yadaiah received B.E. in Electrical and
The authors would like to thank the
comments from the reviewers. Also would like to Electronics Engineering from College of
thank the authorities of JNTU College of Engineering, Osmania University, Hyderabad, India,
Engineering, Anantapur, India and Power Generation in 1988. M.Tech. in Control systems from IIT
Corporation Limited, Hyderabad, India for providing Kharagpur, India in 1991 and Ph.D. in Electrical and
all infrastructural facilities. Electronics Engineering from Jawaharlal Nehru
REFERENCES Technological University, Hyderabad, India in 2000.
[1] IEEE guide for the Interpretation of Gases He received Young Scientist Fellowship (YSF) of
Generated in Oil-Immersed Transformers, Andhra Pradesh State Council for Science and
ANSI/IEEE std.C57.104, 1991. Technology, in 1999. He is currently a Professor in
[2] C.E. Lin, J.M. Ling and C. L. Huang, “An expert Electrical and Electronics Engineering at Jawaharlal
system for transformer fault diagnosis and Nehru Technological University. He is Fellow of
maintenance using dissolved gas analysis” IEEE Institution of Engineers (India), Institution of
Trans. on Power Delivery, Vol.8, No.1, 1993, pp
Electronics & Telecommunications Engineers
231-238.
(India), Member of IEEE, Life Member of Indian
[3] IEEE trial-use guide for the detection of Acoustic
Society of Technical Education, Life Member of
Emissions from partial discharges in oil-
immersed Power Transformers, IEEE std. Systems Society of India. His research includes in
C57.127-2000. the areas of Artificial Neural Networks, Fuzzy Logic,
[4] Y. C. Huang and Hong-Tzer Yang, “Developing a System identification, fault identification, Adaptive
New transformer fault Diagnosis System through Control, Nonlinear Systems and process control.
Evolutionary fuzzy logic”, IEEE Trans. on
Power Delivery, Vol. 12, No.2, 1997, pp 761- N. Ravi received B.E. in Electrical & Electronics
767. Engineering from College of Engineering, Jawaharlal
[5] R. P. Lippmann, “An Introduction to Computing Nehru Technological University Hyderabad, India, in
with Neural Nets”, IEEE ASSP magazine, Vol. 1998 and M.Tech. in Power Electronics from the
4, No.2, 1987, pp 4-22. same College in 2006. He is currently working as an
[6] N. Yadaiah, L.Sivakumar, B.L.Deekshatulu and Assistant Divisional Engineer in Andhra Pradesh
V.Sri Hari Rao, "Neural Network Architectures Power Generation Corporation Ltd., Andhra Pradesh,
for Describing Nonlinear Input-Output Hyderabad, India.
202
3. Design of Fuzzy Controller for a Robot arm: A systematic
Approach
B.Krishna Madhuri
Abstract— less trail and error effort and produce better control
performance.
Most fuzzy controllers have been treated and In literature analytical structures are derived
used as black-box controllers in the sense that their
mainly for fuzzy controllers using Zadeh AND
analytical structures are unknown. Knowing the
explicit structure information will enable one to operator [1]-[5]. Revealing the analytical structure of
insightfully understand how a fuzzy control works. In a fuzzy controller that uses Zadeh AND operator is
the present paper a novel technique is presented for far more difficult even for triangular input fuzzy sets,
deriving input out put relations for the fuzzy which are simplest fuzzy sets because this operator
controller that uses Zadeh AND operator for requires the comparison of membership functions.
symmetrical triangular input fuzzy sets. The robotic
arm control with and without the gravitational effect Recently that is extended to derive the
is simulated to know the efficacy of the fuzzy relationship between input space divisions needed
controller using symmetrical input membership
due to the use of Zadeh AND operator and input
functions.
fuzzy sets [6][7].
Index Terms-- Fuzzy controller, triangular fuzzy In this paper an analytical structure for
sets, robot arm, gravitational torque. symmetrical class of input fuzzy sets is developed.
The robotic arm control is simulated to know the
I. INTRODUCTION efficacy of the fuzzy controller.
Fuzzy controllers are constructed generally via
heuristic approaches as opposed to the mathematical II. DEVELOPMENT OF A GENERAL
approaches exclusively used in conventional control. TECHNIQUE
The fuzzy controllers have been treated and used as A Mamdani type fuzzy controller, which
black box controllers. With out analytical structure employs two input variables, is taken. The two input
information, precise and effective mathematical variables are e (n) and r (n) and system output is y
analysis and design are very difficult to achieve. (n). The scaled error and change of error of y (n) are
Availability of the structure information may lead to
203
E(n) = Ke e(n) = Ke (SP(n)-y(n))
0 , (− , ∞− 1 2 ]
R(n) = Kr r(n) = Kr (e(n)-e(n-1))
R ( +n1 ) 2
Where Ke and Kr are scaling factors and μ = , [− 1 21 , 2 ]
SP (n) is output command signal. E (n) and R (n) are
R j+ 1
2 4
fuzzified by triangular fuzzy sets. The membership 1 [ 1 ∞2] ,
functions for E (n) and R (n) are noted by µ E i, µ E i+1,
µ R j and µ R j+1 respectively.
Over all input space divisions of all the four rules due
The rules are: to Zedah AND operator are shown in Fig. 1.
where K u is the scaling factor and U (n) is the Table I: Rule Firing
output of fuzzy controller.
Analytical structure of the fuzzy controller is
obtained by putting these results into the defuzzifier,
III. ANALYTICAL STRUCTURE FOR
TRIANGULAR MEMBERSHIP FUNCTIONS: and it is tabulated in Table II
Mathematical definitions of the fuzzy sets are
IV. CASE STUDY
A system representing the complete modal for
0 , (− , ∞− 1 5 ] Robot arm control is shown in Fig. 2. with fuzzy PD
controller. The disturbance Tgr is due to gravity.
E ( +n 1) 5 Simulated results are shown in Fig. 3 to Fig. 5.
μE i + 1 = , [− 1 51 , 5 ]
3 0 Fig. 2 Block diagram for the control of robot arm
using fuzzy controller
1 , [ 1 ∞5 ] ,
204
B. Performance Analysis:
From the graphs it can be observed that the rise
timeand settling time for the fuzzy controller using
symmetrical input fuzzy sets are less compared to
rise time and settling time for the conventional
controller. I.e. fuzzy controller gives the fast
response.
C. When arm is moving down:
The effectiveness of controller when the arm is
moving down is shown in Fig. 5. It is observed that
Table II Analytical Structure fuzzy controller response compared to conventional
controller is better in this case also.
V. Conclusions
V. REFERENCES
[1] H. Ying, W. Siler, and J.J. Buckley,”Fuzzy
control theory: A Non-linear case,” Automatica,
Fig. 3 Step response without disturbance vol.26, pp.513-520, 1990.
[2] A. E. Hajjaji and A. Rachid, “Explicit formulas
A. Effect of gravitational torque: for fuzzy controller,” Fuzzy Sets Syst., vol. 62, pp.
Introducing the gravity torque Tgr = 21 sinθ for the 135-141, 1994.
system considered in fig.4., there are two cases to be [3] W. Li, ”Design of a hybrid fuzzy logic
proportional plus conventional integral-derivative
analyzed.
controller,” IEEE Trans. Syst., vol. 6, no. 4, pp. 449-
463, Nov. 1998.
[4] G.K. Mann, B.G. Hu, and R. G. Gosine,
“Analysis of direct action fuzzy PID controller
structures” IEEE Trans. Syst., Man, Cybern, B.
Cybern., vol. 29, pp.371-388, 1999.
[5] Hao Ying, “ A General Technique for Deriving
Analytical Structure of Fuzzy Controllers Using
Fig. 4 Response with disturbance Arbitrary Trapezoidal Input Fuzzy Sets and Zadeh
205
AND Operator”. Automatica, Vol .39, pp. 1171- Stabilizer” Proceedings of IASTED Modeling,
1184, 2003. simulation, and Optimization MSO2004 August 2004
[6] K.A. Gopala Rao, K.R.Sudha, “Analytical
Structure of Three Input Fuzzy PID Power System VI. ACKNOWLEDGEMENTS
Stabilizer with Decoupled Rules”, WSEAS Trans. on I thank my guide Dr. K.A. Gopala Rao, Professor and
CIRCUITS and SYSTEMS, pp.965-970, June 2004. Head, Department of Electrical Engineering for
[7] K.A. Gopala Rao, K.R.Sudha, “Analytical introducing me to this area of research.
Structure of PID Fuzzy Logic Power System
y1(t) A 1 C E
fuzzy gain scheduling of PI controllers have been
proposed to solve such problems in power systems
[4] and [5] that developed different fuzzy rules for
the proportional and integral gains separately. Fuzzy
logic control presents a good tool to deal with
y 2(t) A 2 C E
and settling time with the proposed FGPI controller
are better than the rest.
207
ACE i = ∆Ptie ,i + bi ∆f i ,
(3) Gain scheduling is an effective way of controlling
systems whose dynamics change non-linearly with
operating conditions [4]. It is normally used when the
Where bi is the frequency bias constant, ∆ f i is the
relationship between the system dynamics and
frequency deviation and ∆ P tie ,i is the change in operating conditions are known, and for which a
tie-line power for areai and C is the output matrix [4]. single linear time-invariant model is insufficient. In
this paper, we use this technique to schedule the
parameters of the PI controller according to change
of the new area control error ACE, and ∆ACE, as
depicted in Fig. 3.
In this figure, Kp and Ki are the proportional and Fig.5. Membership functions for FL Controller of
integral gains, respectively. The fuzzy controller (a) ACE, (b) ∆ACE, (c) Kp, Ki
input can be the derivative of e together with the
signal E. The fuzzy controller block is formed by
fuzzification of E, the inference mechanism and
defuzzification. Therefore, Y is a crisp value, and u is
a control signal for the system.
Time(sec)
Fig b. With PI Controller
Fig.6. Membership functions for FGPI Controller Fig 8. e,f shows the responses for Change in
of (a) ACE, (b) ∆ACE, (c) Kp, Ki mechanical power in area1(∆Pm1).(ii)Change in
Membership functions shapes of the error mechanical power in area2(∆Pm2).Change in Tieline
and derivative error and the gains are chosen to be power (∆Ptie).
identical with triangular function for both fuzzy logic
controllers. However, their horizontal axis ranges are
taken different values because of optimizing these
controllers. The membership function sets of FL for
ACE, ∆ACE, Kp and Ki are shown in Fig. 5, while
the ones for FGPI controller are shown in Fig.6.
Defuzzification has also been performed by the
center of gravity method in all studies.
∆f1Simulation study.
5. ∆f11
Simulations
1 were performed using the conventional
PI, Fuzzy Logic (FL) and the proposed FGPI
controllers applied to a two-area interconnected
electrical power system. The same system parameters
given in Tables 2 were used in all controllers for a
comparison. Time(sec)
Two performance criteria were selected in the Fig c. With Fuzzy Logic Controller
Frequency Deviation in area 1
simulation. The frequency deviation graphs were first (∆f1)
plotted with Matlab 7.0-Simulink software. Here, Controller
settling times and overshoots of the frequency Settling Maximum
deviation of the controllers were compared against time(sec) (for Overshoot
each other. The comparison results are provided in 5% band
Table 2 and 3. (HZ)
of the step
change)
Time(sec)
Fig a. Without Controller
209
System performances for all controllers with
steady state error for frequency deviation of area1.
Time(sec)
Fig d. With FGPI Controller 6.Conclusion.
Frequency Deviation in
area 1 (∆f1)
Controller
Steady state error(ess)
Table 2 FGPI -0.000067
FLC -0.00383
Conventional PI -0.00136
Appendix.
210
[6]. Tang KS, Man KF, Chen G, Kwong S. “An optimal
fuzzy PID controller”. IEEE Trans Ind Electron 2001;
48(4):757–65.
References. [7]. CC. Lee,” Fuzzy logic in control systems: Fuzzy logic
controller, parts I and II.” IEEE Trans. Sq’st., Man
Cybern... 20 (2) (1990).
[1]. Demiro¨ren A, Yes¸il E. “Automatic generation
control with fuzzy logic controllers in the power system [8]. El-Sherbiny MK, El-Saady G, Yousef AM.”Efficient
including SMES units”. Electr Power Energy Syst 2004;26: fuzzy logic load-frequency controller”. Energy Convers
291–305. Mgmt 2002; 43:1853–63.
[2]. C¸ am E, Kocaarslan I. “Load frequency control in two [9].Chaturvedi DK, Satsangi PS, Kalra PK. “Load
area power systems using fuzzy logic controller”. Energy frequency control: A generalised neural network
Conversion Manage 2005; 46:233–43. approach”. Electr Power Energy Systems 1999;21:405–15.
[3].Meliopoulos APS, Cokkinides GJ, Bakirtzis [10]. Wang Y, Zhou R, Wen C.” Robust load-frequency
AG.”Load-frequency control service in a deregulated controller design for power systems”. IEE Proc Control
environment”. Decision Support Syst 1999; 24:243–50. 1993;140(1):11–16.
[4]. Talaq J, Al-Basri F. “Adaptive fuzzy gain scheduling
for load frequency control”. IEEE Trans Power syst 1999;
14(1):145–50.
Abstract—This paper presents an enhanced genetic with normal genetic algorithm and enhanced genetic
algorithm (EGA) for the solution of the optimal algorithm are shown.
power flow (OPF) with both continuous and discrete
control variables. The continuous control variables NOMENCLATURE
modeled are unit active power outputs and
θ Bus voltage angle vector.
generator-bus voltage magnitudes, while the discrete
ones are transformer-tap settings and switchable UL Load (PQ) bus voltage magnitude vector.
shunt devices. A number of functional operating
constraints, such as branch flow limits,load bus PG Unit active power output vector.
voltage magnitude limits, and generator reactive UG Generation (PV) bus voltage magnitude
capabilities, are included as penalties in the GA
fitness function (FF). Numerical results on ieee test vector.
system are presented and comparision of results
t Transformer tap settings vector.
211
bsh Bus shunt admittance vector. can be solved using a Newton-type algorithm. In
Newton OPF [6],the inequality constraints are added
X System state vector. as quadratic penalty terms to the problem objective,
multiplied by appropriate penalty multipliers.
U System control vector. Interior point (IP) methods, convert the inequality
constraints to equalities by the introduction of
A hat above vectors and denotes that the entry nonnegative slack variables. A logarithmic barrier
corresponding to the slack bus is missing. For function of the slack variables is then added to the
objective function, multiplied by a barrier
simplicity of notation, it is assumed that there is parameter, which is gradually reduced to zero
only one generating unit connected on a bus. during the solution process. The unlimited point
algorithm uses a transformation of the slack and
I. INTRODUCTION dual variables of the inequality constraints which
converts the OPF problem KKT conditions to a set
Since its introduction as network constrained
of nonlinear equations, thus avoiding the heuristic
economic dispatch by Carpentier and its definition rules for barrier parameter reduction required by IP
as optimal power flow (OPF) by Dommel and meth ods.
Tinney , the OPF problem has been the subject of
OPF programs based on mathematical
intensive research. The OPF optimizes a power
programming approaches are used daily to solve
system operating objective function (such as the
very large OPF problems.However, they are not
operating cost of thermal resources) while satisfying
guaranteed to converge to the global optimum of the
a set of system operating constraints, including
general nonconvex OPF problem, although there
constraints dictated by the electric network. OPF has
exists some empirical evidence on the uniqueness of
been widely used in power system operation and
the OPF solution within the domain of interest To
planning . After the electricity sector restructuring,
avoid the prohibitive computational requirements of
OPF has been used to assess the spatial variation of
mixed-integer programming, discrete control
electricity prices and as a congestion management
variables are initially treated as continuous, and
and pricing tool .In its most general formulation, the
post-processing discretization logic is subsequently
OPF is a nonlinear, nonconvex, large-scale, static
applied Whereas the effects of discretization on
optimization problem with both continuous and
load tap changing transformers are small and usually
discrete control variables. Even in the absence of
negligible, the rounding of switchable shunt devices
nonconvex unit operating cost functions, unit
may lead to voltage infeasibility, especially when
prohibited operating zones, and discrete control
the discrete VAR steps are large, and requires
variables, the OPF problem is no convex due o the
special logic.The handling of nonconvex OPF
existence of the nonlinear (AC) power flow equality
objective functions, as well as the unit prohibited
constraints. The presence of discrete control
operating zones, also present problems to
variables, such as switchable shunt devices,
mathematical programming OPF approaches.
transformer tap positions, and phase shifters, further
complicates the problem solution. Recent attempts to overcome the limitations
of the mathematical programming approaches and
The literature on OPF is vast, and [1] presents
the major contributions in this area. Mathematical genetic algorithms (GAs) [7], [8].
programming approaches, such as nonlinear
programming (NLP), quadratic programming (QP) In [7], a simple genetic algorithm (SGA) is
[2], [3], and linear programming (LP) [4],[5], have used for OPF solution. The control variables
been used for the solution of the OPF problem. modeled are generator active power outputs and
Some methods, instead of solving the original voltages, shunt devices, and transformer taps.
problem, solve the problem’s Karush–Kuhn–Tucker Branch flow, reactive generation, and voltage
(KKT) optimality conditions. For equality- magnitude constraints are treated as quadratic
constrained optimization problems, the KKT
penalty terms in the GAfitness function (FF).To
conditions are a set of nonlinear equations, which
212
keep the Ga chromosome size small, only a 4- The OPF problem can be formulated as a
bitchromosome area is used for the encoding of each mathematical optimization problem as follows:
control variable. A sequential GA solution scheme
is employed to achieve acceptable control variable Minƒ(x,u) (1)
resolution. Test results on the IEEE 30-bus system,
comprising 25 control variables, are presented. S.t.g(x,u) = 0 (2)
Assuming an initial random population produced Until now, many crossover schemes, such as single
and evaluated, genetic evolution takes place by point,multipoint, or uniform crossover have been
means of three basic genetic operators:
proposed in the literature. Uniform crossover [9] as
1) Parent selection; been used in our implementation. While crossover is
Parent selection is a simple procedure whereby two responsible for the injection of new information.
chromosomes are selected from the parent With a small probability, random bits of the
population based on their fitness value. Solutions offspring chromosomes flip from 0 to 1 and vice
with high fitness values have a high probability
versa and giveconstraints . new characteristics that
Of contributing new offspring to the next do not exist in the parent population . In our
generation. The selection rule used in our approach
approach, the mutation operator is applied with a
is a simple roulette-wheel selection [9]. Crossover is
an extremely important operator for the GA.It is relatively small Probability (0.0001-0.001) to every
responsible for the structure recombination bit of the
(information exchange between mating chromosome.
214
The FF evaluation and genetic evolution take
part in an iterative procedure, which ends when a
maximum number of generations is reached, as 2) discrete controls taking values ui 1, ui 2….., ui m
1) continuous controls taking values in the Fi( PGi ) fuel cost function of unit (in our case, a
interval [u ,u
i
min
i ]
max
ui = uimin +( u i
max
_− uimin).K/(2Nui -1) (7) quadratic function);
This operator is applied only to the best Fig. 6. Gene max-min operator.
chromosome (elite) of every generation [8].In
2) Gene Copy Operator (GCO): This operator
addition to the above advanced features, which are
randomly selects one gene in a chromosome and
called “advanced” despite their wide use in most
with equal probability copies its value to the
recent GA implementations to distinguish between
predecessor or the successor gene of the same
the SGA and our EGA, operators specific to the
control type, as shown in Fig. 4. This operator
OPF problem have been added.
has been introduced in order to force consecutive
All problem-specific operators introduce
random modification to all chromosomes of a new controls (e.g., identical units on the same bus) to
generation. If the modified chromosome proves to operate at the same output level.
have better fitness, it replaces the original one in the
new population. Otherwise, the original 3) Gene Inverse Operator (GIO): This operator
chromosome is retained in the new population. All acts like a sophisticated mutation operator. It
problem-specific operators are applied with a randomly selects one gene in a chromosome and
probability of 0.2. The following problem-specific inverses its bit-values from one to zero and vice
operators have been used. versa, as shown in Fig. 5. The GIO searches for bit-
structures of improved performance, exploits
1) Gene Swap Operator (GSO): This operator newareas of the search space far away from the
randomly selects two genes in a chromosome and current solution, and retains the diversity of the
swaps their values, as shown in Fig. 3. This operator population.
swaps the active power output of two units, the
voltage magnitude of two generation buses, etc. 4) Gene Max-Min Operator (GMMO): The
Swapping among different types of control variables GMMO tries to identify binding control variable
is not allowed. upper/lower limit constraints.It selects a random
gene in a chromosome and, with the same
217
probability (0.5), fills its area with 1 s or 0 s, as transformer-tap settings, and nine bus shunt
shown in Fig. 6. admittances. The input data for IEEE 30-Bus
System are given in appendix ‘A’ .The gene length
D. Enhanced Genetic Algorithm (EGA) for unit power outputs is 12 bits and for generator
voltage magnitudes is 8 bits. They are both treated
In the EGA, shown in Fig. 7, after the application of
as continuous controls. The transformer-tap settings
the basic genetic operators (parent selection,
can take 17 discrete values (each one is encoded
crossover, and mutation) the advanced and problem-
using 5 bits): the lower and Fig. 8. FF comparison
specific operators are applied to produce the new
for IEEE 30-bus system. upper limits are 0.9 p.u.
generation. All chromosomes in the initial
and 1.1 p.u., respectively, and the step size is 0.0125
population are created at random (every bit in the
p.u. The bus shunt admittances can take six discrete
chromosome has equal probability of being
values (each one is encoded using 3 bits): the lower
switched ON or OFF). Due to the decoding process
and upper limits are 0.0 p.u. and 0.05 p.u.,
selected [(7) and (8)], the corresponding control
respectively, and the step is 0.01 p.u. (on system
variables of the initial population satisfy their
MVA basis). The GA population size is taken equal
upper–lower bound or discrete value constraints (4).
to 60, the maximum number of generations is 200,
However, the initial population candidate and crossover and mutation are applied with initial
solutionsmay not satisfy the functional operating probability 0.9 and 0.001, respectively. Two sets of
constraints(3) or even the load flow constraints (2) 20 test runs were performed; the first (SGA) with
since the random,within limits, selection of the only the basic GA operators and the second (EGA)
control variables may lead to load flow divergence with all operators, including advanced and problem-
(as already discussed in Section I V-B). Population specific operators. The FF evolution of the best of
statistics computed for the new generation include these runs is shown in
maximum, minimum, and average fitness values and
the 90% percentile.
system
Fig. 7. Enhanced genetic algorithm (EGA). The best and worst solutions of the second set of 20
runs (EGA) are shown in Table I. The operating
V. TEST RESULTS
costs of the best and worst solutions are 802.73$/h
In this section, the proposed EGA solution of the and 802.34 $/h, respectively, The differences
OPF is evaluated using A. IEEE 30-Bus System between the values of the control variables in the
best and worst solutions are significant. The
It has 41-branch system . It has a total of 24 control operating cost of all EGA-OPF solutions is slightly
variables as follows: five unit active power outputs,
six generator-bus voltage magnitudes, four
218
less as shown in Table below [1] J. A. Momoh, M. E. El-Hawary, and R. Adapa,
“A review of selected optimal power flow literature
to 1993,” IEEE Trans. Power Syst., pt. I and II, vol.
14, pp. 96–111, Feb. 1999.
REFERENCES
219
using genetic algorithm,” Elec. Power Syst. Res.,
APPENDIX ‘A’
No.of buses : 30
No.of generators : 6
Bus data:
Bus Type Pgen Qgen Pload Qload Vspecified Qmin Qmax Yshunt
220
6 2 0.15 5
6 P-Q 0 0 0 0.3 1 0 0 0
7 P-Q 0 0 0.22 0 1 0 0
8 0
9 P-Q 0 0 0 0 1 0 0 0
20 P-Q 0 0 0.02 0 1 0 0
2 0
22 P-Q 0 0 0 0.06 1 0 0
7 0
23 P-Q 0 0 0.03 0 1 0 0
2 0
221
24 P-Q 0 0 0.08 0.02 1 0 0
7 3 0.04
25 P-Q 0 0 0 0 1 0 0 0
26 P-Q 0 0 0.03 0 1 0 0
5 0
27 P-Q 0 0 0 0.00 1 0 0
9 0
28 P-Q 0 0 0 0.01 1 0 0
9 0
29 P-Q 0 0 0.02 1 0 0
4 0
30 P-Q 0 0 0.10 1 0 0
6 0
1 0.5 2.0
2 0.2 0.8
5 0.15 0.5
8 0.1 0.35
11 0.1 0.3
3 0.2 0.8
222
Generator a b c
No
1 0 2 0.00375
2 0 1.75 0.0175
3 0 1 0.0625
4 0 3.25 0.002075
5 0 3 0.025
6 0 3 0.025
Line Data :
13 9 11 0 0.208 0 1 0.65
14 9 10 0 0.11 0 1 0.65
16 12 13 0 0.1400 0 1 0.65
224