Sie sind auf Seite 1von 6

BULETINUL INSTITUTULUI POLITEHNIC IAI

TOMUL L (LIV), FASC. 5, 2004


ELECTROTEHNIC, ENERGETIC, ELECTRONIC

ARTIFICIAL INTELLIGENCE TECHNIQUES IN OPTIMAL


REACTIVE DISPATCH. SOLUTIONS USING GENETIC
ALGHORITMS
BY
*D. BIC, *D. SARCHIZ and *VASARHELY I.

Abstract. Genetic algorithms are becoming popular to solve the optimization


problems in different fields of applications mainly because of its robustness in
finding optimal solution and ability to provide near optimal solution close to
global minimum.
In the present paper, a genetic algorithm to determine the optimal size of the
capacitors under different load conditions is implemented. The objective is to
minimize the energy losses under voltage quality conditions. The voltage
constraint is formulated as a penalty function to the objective function. To test the
proposed genetic algorithm, an power system model has been taken as an
example.
Keywords: power system, artificial intelligence, genetic algorithm, capacitors
optimization.

1. Introduction

Artificial Intelligence (AI) techniques, that is a new branch of computer


science, have in recent years emerged as a complement tool to mathematical
approaches within power systems. Today the main AI used in power system
applications are those utilizing logic and knowledge representations of Expert
Systems (ES), Fuzzy Systems, Artificial Neural Networks (ANN) and more
recently, Genetic Algorithms (GA). The Optimal Reactive Dispatch (ORD) is a
particular case of a more general problem that is the Optimal Power Flow and
consists in determining the state of an electric power system that optimize a
given objective function and satisfies a set of physical and operating constraints.
The ORD has grown in importance for many reason: the need for most efficient
operations of power systems has increased, the losses of the power systems can
be reduced by optimizing the reactive power flow and the voltage is considered
one of the most important parameters of power quality.

2. Some aspects about GA

Genetic algorithm (GA) is a global search technique based on mechanism


of natural selection and genetics. It is a general-purpose optimization algorithm
D. BIC, D. SARCHIZ and VASARHELY I.

that is distinguished from conventional optimization by the use of the concepts


of populations genetics to guide the optimization search. A GA operates on a
random population of individuals each representing to the problem at hand. In
recent years, GA has gained popularity for its easy searching process, global
optimality, independence of searching space and probabilistic nature. Instead of
point-to-point search, GA searches from population to population.
For hard optimization problems, GA are a superior methods by providing
simultaneous both exploitation of the best solutions and exploration of the
search space. The advantages of GA over traditional optimizations techniques
are [4,7]: i) GA needs only rough information of the objective function an place
no restrictions such as differentiability or convexity on the objective function; ii)
The method works with a set of solutions from one generation to the next and
not a single solution, thus making it less likely to converge at a minimum
locally; iii) The solutions developed are randomly based on the probability rate
of the genetic operators such as mutation and crossover. The initial solutions
thus would not dictate the search direction of a GA.
In general, a GA performs a multidirectional search by maintaining a
population of potential solutions and encourages information formation and
exchange between these directions. This populations undergoes a simulated
evolution: at each generation, the relatively good solutions reproduce, while
the relatively bad solutions die. To evaluate the quality of a solution, a special
function is used , playing the role of the environment. In most case, this function
is the function to optimize. To make the transition from the current population to
the next population, three processes are used: selection, crossover and mutation.

Initial Fitness
population evaluation Stop?
Start
no

New Best
population SELECTION cromozome
sol

CROSSOVER solution
New
population

MUTATION

Fig. 1.- Genetic algorithm - basic cycle.


BUL. INST. POLIT. IAI, TOMUL L (LIV), FASC. 5

3. GA applications in power systems ORD. Mathematic model

The objective of the problem is to find number, location and discrete size
of capacitor to be placed in the system in such a way that the total cost of energy
loss, including the cost of capacitor is minimum, under energy quality
constraints.
Objective function: the objective function for optimization can be stated
mathematically as [1, 2]:
ncap
Minimize: F (Q) = K e PT + K c (Qi ) (1)
i =1
where P is the active power loss at the power system with a time duration T,
K c (Qi ) is the cost of the ith capacitor of rating Qi , ncap the number of the
candidate locations for capacitor placement and K e is the energy cost of losses.
One of the main objectives of compensation problem is to place the minimum
number of capacitor so that installation cost could be minimum. Hence it will be
worth-while, if the installation cost of the capacitor could be incorporated in the
capacitor cost calculation itself. Both the installation cost and marginal cost of
capacitor has been considered separately. The capacitor cost is considered as the
linear function of its size:
K c (Qi ) = K ci + K c Qi (2)
where K ci is installation cost of the ith capacitor.
Constraints: Two types of constraints are imposed:
a) Reactive power constraints:
Qimin Qi Qimax i = 1,..., ncap

b) Voltage constraints: The voltage constraints will be formulated as a


penalty function of objective function. The constrained objective function is:
ncap
F (Q) = K e PT + K c (Qi ) + T Vmin
l
Vs (3)
i =1
l
where Vmin is the vector containing the minimum voltage limit for each load
bus, Vs is the corresponding vector of the load bus voltage and is the vector
of constant multiplier which determine the importance of each voltage constraint
in the objective function. A higher value of elements can cause premature
convergence while with lower value of vector elements, the GA may create
individuals that violate the constraint but are rated than those do not violate the
constraints. Hence suitable value of , satisfying the constraint limits and also
giving best possible result could only be found out by trial and error.
D. BIC, D. SARCHIZ and VASARHELY I.

4. Implementation of GA model. Test results

In the GA, the fitness function is defined as a nonnegative function of


merit to be maximized which is directly associated with the objective function
(1) or (3).We consider the inverse function of F as the fitness function, then it
will the both conditions (nonnegative and to be maximized). Therefore for the
given problem the fitness function will be defined as K/F, where K is a constant
[ ]
multiplier. A chromosome from the population is formed as Q1Q2...Qncap where
Qi represent the binary string of nb =6 bits representing the value of the
compensation reactive power from the ith candidate location, allowing that the
value of variable to be optimized (reactive power) vary for the given
[ ]
Qmin, Qmax , taking discrete values by:
Xi
Qi = Qmin + (Qmax Qmin ) (4)
2 1
nb

where X i is the value of the binary string Qi in the base 10. The maximum
number for the generations for the problem is 40.The parameters of the objective
function are: K e = 60[$ / MWh] , K ci = 100[$] , K c = 0.01[$ / MVAr ] and
= 1 .Thus as convergence criteria will be used the difference between average
fitness of entire population calculated in the last two consecutive generation
( = 10 7 ).
The main steps of algorithm of the proposed method are:
1. Form initial chromosomes equal to the population size as: [Q1Q2...Qncap ];
2. Evaluation of the fitness for each chromosome:
i) Convert each binary substring Q in string into decimal number representing
the size of the reactive power;
ii) Assign this reactive powers for the given location and carry out the power
flow calculation and to find out : power losses and nodal voltage.
iii) Calculate the objective function F with (1) then the fitness K/F.
3. Find the best fitted chromosome and send it to the solution vector and
calculate the mean of whole population fitness.
4. Set the generation number=1.
5. i) Select a number of individuals from the current population in proportion
with the selection rate, selection based on the rank method;
ii) Perform crossover and generate offspring;
iii)Mutate these offspring based mutation probability.
6. Calculate fitness for each offspring as in step 2 and step 3.
7. Combine the old population and new population to create a single population
best of both. For example, at t-th generation, total population size is NI(old
population).In step 5 a set of new population is obtained NIc in proportion
BUL. INST. POLIT. IAI, TOMUL L (LIV), FASC. 5

with the selection rate. When combining these old and new population,
results a population with NI+ NIc size, from wich will survive in the t+1-th
generation only NI individuals- the best fitted.
8. Repeat step 5 to step 7 till the solution converges or a maximum number of
generation is reached.
A test 6 buses - power system is given, represented in figure 2. The reactive
power compensation will be calculated for load buses 2, 4 and 5.The results
obtained after running the optimization program based on a GA are
represented in figures 3, 4 and 5.

1-NG 2 3-NG

C2

5 4

C5
C4

6 -NE

a) b)
Fig. 2.- The 6 buses- test power system: single diagram (a) and daily power features P(t) and
Q(t) on load buses (b).

a) b)
Fig 3. Daily active power losses P (t) : initial steady state and optimal steady state (a),
Instantaneous compensation values Q (MVAr) in the load buses for optimal rating (b).
D. BIC, D. SARCHIZ and VASARHELY I.

Appendix

* Petru Maior University of Tg.Mure, Department of Electrical Engineering,


Tg.Mure, Romania.

REFERENCES
1. Das, D., Reactive power compensation for radial distribution networks using genetic
algorithm. Electrical Power and Energy System, Elesevier Ltd, 573-581 (2001).
2. Sarchiz, D., Applications of Genetic Algorithms in optimization in Power Systems. Proc.of
Conference Inter-Ing 2003, Petru Maior University of Tg.Mure, 203 -208 (2003).
3. Eremia, M,. s.a., Tehnici de inteligen artificial.Concepte i aplicaii n sistemele
electroenergetice .Editura AGIR, Bucureti, 2001.
4. Bansal, R.C., Bhatti, T.S., Kothari, D.P., Artificial Intelligence for reactive power/voltage
control in power system: A review. International Journal of Power and Energy System,
Vol.23. No.2, 81-88 ( 2003).
5. Buiu, C., Evolutionary computing. From natural to intelligent machines. Proc. Advanced
Technologies in power systems: FACTS and I.A., Bucharest, 407-426 ( 2003).
6. Bic, D, Contribuii la optimizarea regimurilor de funcionare i stabilitatea de tensiune a
sistemelor electroenergetice.Tez de Doctorat, Universitatea Politehnica Bucureti, 2003.
7. Bic, D., Bucur, D., Sistemele expert-opiuni atractive n studiul sistemelor electroenergetice.
Buletinul tiinific al Universitii Petru Maior din Tg.Mure, 45-51(2003).
8. Dumitrache, I. s.a, Tehnici de conducere intelligent a proceselor.Universitatea Politehnica
Bucureti, 1993.

TEHNICI DE INTELIGEN ARTIFICIAL N OPTIMIZAREA PUTERII REACTIVE. SOLUTII


UTILIZND ALGORITMI GENETICI.
(Rezumat)

Algoritmii genetici au devenit tehnici populare, utilizate n problemele de optimizare din


diferite domenii i aplicaii, datorit robusteii n gsirea soluiei optime i capacitii lor de a
furniza o soluie optimal aproape de minimul global. Lucrarea propune o soluie de aplicare a
algoritmilor genetici n determinarea valorilor optime ale capacitilor de compensare ntr-un
sistem electroenergetic, funcionnd sub ncrcri variabile. Obiectivul este de a minimiza
pierderile de putere activ cu respectarea restriciilor privind nivelul de tensiune n noduri.
Restriciile privind tensiunea sunt introduse n funcia obiectiv printr-un factor de penalizare.

Das könnte Ihnen auch gefallen