International Conference and Workshop on Emerging Trends in Technology (ICWET 2010) – TCET, Mumbai, India

Goal Oriented Ranking Scheme

A R Khaparde A Mokhade P D Dahiwale
Dept of Computer Science And Dept. Electronic and Computer Dept of Information Technology,
Engineering, NYSS college of Science, Visvesvararya National NYSS college of Engineering and
Engineering and Research Institute of Technology, Research
09923294516, India 09422457645,Nagpur,India 9970380078, India
khaparde.amit@gmail.com asmokhade@cse.vnit.ac.in raramit2@gmail.com

ABSTRACT multi objective EA based on non-domination on a constrained

Evolutionary algorithms (EAs) are always be a good choice to
solve multi-objective problems (MLPs). EA use ranking schemes
to solve MLP‟s. Most of ranking schemes use only objective
values to rank the solution where as some use the condition in
which the problem arises to rank the solutions. In this paper we
have proposed one ranking method which use situation as well as
objective values to rank the solutions. This method is currently
being tested on standard single and multi objective problems, the
early results are encouraging.
Categories and Subject Descriptors
I. [Computing Methodology]: Artificial Intelligence–learning,
Parameter learning
General Terms
Algorithms, Performance, Design.
Keywords
Multi objective problem, Goal attainment method, Decision maker
(DM), ID , dM, rank.
1 INTRODUCTION
During the last few decades Evolutionary Algorithms (EAs) have
proved to become an important tool to solve difficult search and
optimization problems. Most real-world problems have constrained
and EAs do not have a efficient and generic constraint handling
techniques.
The constrained optimization problem may be handled as a multi
objective optimization problem as indicated by Coello Coello [2],
Michalewicz [9] and Fonseca and Fleming [5]. Furthermore,
optimization problem leads to a search of the best compromises of
the objective value and constraint satisfaction, This whole set of
solutions is usually not interesting since it is the optimal and
feasible solution that is searched. Therefore it will not be efficient
to directly apply a multi objective EA on a constrained problem;
still the idea to handle the constrained problem with some variant
of a multi objective EA is interesting.
One of the most crucial steps in a multi objective EA is how to
rank individuals. In this paper an alternative ranking scheme for
the constrained single and multi objective problem is introduced.
This ranking scheme is generic and no new parameters are
introduced. The ideas of the ranking scheme are borrowed from the
non-domination ranking for multiple objectives by Goldberg
[6],Ander‟s ranking method [1] and Fonseca and Fleming ranking
approach [4].The paper first defines the constrained optimization
problem, and !hereafter the proposed method is presented in more
detail.
2 THE CONSTRAINT OPTIMIZATION
PROBLEM
The constraint optimization problem or non -linear programming
with „k‟ inequality constraint and „m‟ equality constraint is
formulated as:
(1)
Such that (2)
(3)
, is a vector of „n‟ design variables such

EAs based on non-dominated sorting for multiobjective problems that . The search space S is define as an n-dimensional
have received large interest during the past decades. Therefore it rectangle by upper and lower for design variables,
seems natural to look upon the constrained optimization problem the feasible region is the region of S
as a multi objective problem. One of the interesting constraints
for which the inequality and equality constraint are satisfied. The
handling method based on non-domination is presented by Deb et
al. [7], Multi objective approaches of constrained problems based optimal solution is denoted by . A constraint is said to be active
on Shaffers VEGA [11] is found in [12] [10].To directly apply a at point if . By default all equality constraint are
Permission to make digital or hard copies of all or part of this work for active at all point of feasible space.
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that 2.1 Incorporate the Decision Making in
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
Problem
requires prior specific permission and/or a fee. Goal attainment method [11] on the basis of that decision maker
(DM) can decide the goal for the given objective in particular
ICWET‟10, February 26–27, 2010, Mumbai, Maharashtra, India.
problem. The equation 1 can be written as
Copyright 2010 ACM 978-1-60558-812-4…$10.00.

505
 International Conference and Workshop on Emerging Trends in Technology (ICWET 2010) – TCET, Mumbai, India

It is depending upon the DM that to what extent he wants to to remaining components or if the remaining components of do
achieve the goal. This can be obtain by using the following not meets the goal. i.e
equation

is the objective function and is the goal associate with

objective function , „w‟ is the weight assign to the objective,
Case B
and .
if . Satisfies none of the goal. Then,
3. PRAPOSED APPROCH OF
is preferable on if and only if dominates .
CONSTRAINT OPTIMIZATION
The new ranking scheme is use to formulate a scalar value function
that is use to rank individual in current population. Then selection,
crossover, mutation and reinsertion are used in a standard manner Case C
for a real coded GA in this paper. Our main focus here is to define if . Meets all of the goals, which means
a new ranking scheme that it is a satisfactory though not necessarily optimal, solution.
The approach is based on the following criteria.
is preferable to if and only if it dominates or is not
If no feasible individual exists in the current population,
the search should be directed towards the feasible region. satisfactory .
If majority of individuals in population are feasible, the
search should be directed towards the individuals who On the basis of above criteria found out the ID and dM values for
satisfy the goal. all individual in population.
A feasible individual closer to the goal value always Once the ID and dM for both and values are found,
better than a feasible individual further from the goal.
assign the and on basis of dM value . Lower the dM
value higher the rank. At this stage each individual has and
3.1 Procedure to find ID and dM value of
. P is the size of population . N is number of feasible
rank1
Suppose there is ‘q’ number of objectives and at certain solutions. Now new objective function ( ) is formulated as
instant assume the two q-dimensional objective vectors
(6)
are and is said to be better
than iv. 4 An illustrative example
if ( [ai]< [ai] for all i=1…….q . In this section a simple example is solve by using proposed
method, main purpose of this section to show the important effect
The dominates more number of objective values of the method not to solve the problem. First the result of an actual
of . search is presented. Then the imposed search direction is discussed
If both and dominate equal number of objective values or all with the help of hypothetical population.
The problem is as follow. A quadratic function is to be minimized
the objective values of both and are equal than they are said
and the feasible solution are constrained by three circles .problem
to be non-dominating individuals. is stated as
3.2 Procedure to find ID and dM value of
rank2
Again we are considering the and
and goal vector is . Also
consider that is such that it meets a number from 1 to ‘q-
k’ number of specified goal. There are three cases.
Case A
: :
For the unconstraint problem the optimal solution is = [-
If ( ) Meets goals from 1....k and it will 0.1679,-0.01329] the population size set to 10 and , the maximum
evaluation is set to 50.the crossover and mutation probability set to
preferable to , If it dominates with respect to its k+1........ q.
be 0.6 and 1/n respectively . The initial and final generation
For the case where all of the k+1....q components are equal according to eq(6) is given in figure 1.
to . is still be preferable to , if it dominates with respect

506
 International Conference and Workshop on Emerging Trends in Technology (ICWET 2010) – TCET, Mumbai, India
Figure1: the numbering shows the ranking at starting of
evolution. And dots show the population in final evaluation.
At start it was found that was generated between the range of -
4 to 7 and was generated between the range of -2 to 5. The
individuals were scattered in between above ranges, and no
solution was feasible solution. In first evaluation the search was
moving towards to found the feasible solution but here the range of
both variables i,e were -2 to 2.5 and -1 to 4.5, respectively.
At and still there was no feasible solution but
range of solution space was moved toward feasible solution ,
ranges was 0 to 4,-1.5 to 2 and 0 to 4.5 ,-0.5 to 2 , respectively. up
to this stage the eq(6) assign the final rank of all the individuals
equals to , because there was no feasible solution , but at
evaluation the feasible solution was found so final
ranking took both and in an account . As the number
of feasible individuals increases search was moved towards to
individuals which would satisfy the goal i,e the goal value of
individuals was more closed to goal value (assign by DM). From
the evaluation 5 onwards the number of feasible solution were
increased hence the search moved to solution region which
satisfied all the constraint and the more closer to goal, in order to
study the behavior during this situation we taking the cases when
the number feasible solution are 15% of total population and when
the feasible solutions are 45% of population.
Figure 2
507
Figure 3
Circle shows the feasible solution and diamond shows the rest
of individuals in population .square represent the most better
individual (final_rank =0)
From the figure 2 the number of feasible solutions is less hence
the weight of is less but still it moving towards the region
where more number of feasible individuals can found which has
values closer to the goal (by DM), and in figure 3 there are almost
double in number of feasible individuals compare to figure 2. As
the number of evaluation increases the number of feasible
individual increases and more better solution will found. The
number of feasible individuals are found during evaluation is
represented in the Figure 3. Figure 1 satisfy the criteria 1 i,e start
with random population and finding the feasible solution . Figure 2
satisfy the criteria 2. i.e. here have more than one feasible solution,
finding feasible solution closer to goal value(By DM).figure 3
satisfy the criteria 3 i,e the individual closer to goal is better than
one which is away from goal.
.
No. of
feasible
solution
s
No. of evaluations
Figure 4
5. CONCLUSION
A general ranking scheme without problem specific extra
parameters for constrained optimization problem has been
presented. The early results are encouraging. the solution of
problem totally depends upon the goal rank and objective rank
.This scheme will give better results if it is used with more advance
genetic algorithms.
6. FUTURE WORIK
Our main purpose here is to proposed the new ranking
scheme , but future work is to check the ranking scheme
with standard test problems which are found in the theory .
 International Conference and Workshop on Emerging Trends in Technology (ICWET 2010) – TCET, Mumbai, India

REFERENCES [8] Michalewicz, Z., 1995, “A Survey of Constraint Handling

Techniques in volutionary Computation Methods”, In J. R.
[1].Anders Angantyr, Johan Andersson, Jan-Olov McDonnell, R. G. Reynolds and D.
Aidanpaa.,Constrained Optimization based on a Multiobjective B. Fogel (Eds.), Proceedings ofthe 4“ Annual Conference on
Evolutionary Algorithm Evolutionary Programming, MIT Press, Cambridge, MA, pp. 135-
[2][ Coello Coello, C. A., 1999, “A Survey of Constraint Handling 155,
Techniques used with Evolutionary Algorithms”, Technical Report [9]. M.M. Raghuwanshi and O.G. Kakde “Survey on
Lania- 1566 RI-99-04, Laboratorio Nacional de Informltica multiobjective evolutionary and real coded genetic
Avanzada, Xalapa algorithms”.
[3]. Floyd w gembicki and yacov y hames. “Approach to
performance and sensitivity multiobjective optimization :goal [10].Parmee; I; C. and Purchase; G., 1994,‟ “The development of a
attainment method “ . directed genetic search technique for heavily constrained search
[4]. Fonseca, C .M. and Fleming P.J.,1993,”Genetic Algorithms , spaces!‟, In I. C. Parmee (Ed.), Proceedings of the Conference on
Discussion and Generalization ”, in S .Forrest(Ed.),proceeding of Adaptive Computing in Engineering Design and Control,
international conference on Genetic Algorithm ,Morgan Kaufmann University of Plymouth, Plymouth, pp. 91- 102.
,San Mateo ,CA, pp 416-423.
[5] Fonseca, C. M. and Fleming, P. J, 1995, “Multiobjective [11] Schaffer, J. D., 1985, “Multiple objective optimization with
Optimization and Multiple Constraint Handling with Evolutionary vector evaluated genetic algorithms”, In J. J. Grefensttete (Ed.)
Algorithms 1: A Unified Formulation”, Research Report 564, Proceedings of the I” Int. Coilference on Genetic Algorithms,
University of Sheffield, Sheffield. Hillsdale, New Jersey: Lawrence Erlbaum Associates, pp. 93-100.
[6]. Goldberg,D.E ,- 1989,Genetic Algorithm in
search,optimization and machineLearning , Addision-wesely , [12] Surry, P., Radcliffe, N., Boyd, I., 1995, “A multiobjective
Reading. approach to constrained optimization of
gas supply networks”, In T. C. Fogarty (Ed.), Proceedings of the
[7] kalyanmoy Deb, Amrit Pratap, Sameer Agrawal, T meyarivan AISB-95 Workshop on Evolutionary Computing, Springer,
.(KANGAL).,The Fast and Elitist multi objective Genetic Sheffield, pp. 166-1 80:
algorithm: NSGAII

508