While searching for an optimum solution using a genetic algorithm, it is always critical to properly balance the adequate exploration of the search space during earlier generations, and putting the required selective pressure to find the optimum during later generations. This change from more explorative search to close selective search depends on the problem and is preferably done adaptively. Common approaches to solve this are fitness scaling, ranking of the chromosomes, tournament selection etc. Other proposals are to adaptively change the probabilities of crossover and mutation operations as the genetic search progresses. Our proposal is to fine tune this by restricting the choice of partners for crossover over generations. In real life, marriages (crossovers) occur between two individuals of similar status in society, only when they are mature and usually from neighboring localities. A similar principle is extended to selecting cross-over partners in GA. In the proposed strategy, the probability of cross-over is higher when their rank in the whole population is close and they are mature. At an early stage, restricting the crossover to chromosomes of similar rank would lead to bad exploration. The probability function for selecting partners depending on their ranks changes with advancing generations, so that the effect is negligible in the beginning. At a later stage the effect is accentuated so as to be able to fine tune good chromosomes to achieve fast convergence and reach optimum values. The scheme is not centralized like the elitist approach. The case of restricting the crossover to partners of mature age was separately studied. The effectiveness of this new method is tried on problems of maximizing complex multimodal functions. The results are compared with the standard genetic algorithm (SGA) and SGA with "linear fitness scaling". Results show that our strategy is superior in terms of probability of hitting the maximum value as well as the speed of finding the maximum.
[1]
J. David Schaffer,et al.
An Adaptive Crossover Distribution Mechanism for Genetic Algorithms
,
1987,
ICGA.
[2]
L. Darrell Whitley,et al.
The GENITOR Algorithm and Selection Pressure: Why Rank-Based Allocation of Reproductive Trials is Best
,
1989,
ICGA.
[3]
Larry J. Eshelman,et al.
Preventing Premature Convergence in Genetic Algorithms by Preventing Incest
,
1991,
ICGA.
[4]
M. Kirkpatrick,et al.
The evolution of mating preferences and the paradox of the lek
,
1991,
Nature.
[5]
Lalit M. Patnaik,et al.
Adaptive probabilities of crossover and mutation in genetic algorithms
,
1994,
IEEE Trans. Syst. Man Cybern..
[6]
Edmund M. A. Ronald,et al.
When Selection Meets Seduction
,
1995,
ICGA.