Flexible Genetic Algorithm for Complex Optimization Problems

Abstract

In this work we propose to specify, describe and test a variant of a more powerful and flexible genetic algorithm that could be better suitable to tackle complex optimization problems such as in dynamic, stochastic or robust optimization. Our main goal is to give a new strong tool more efficient in terms of both solution quality and time processing for complex NP-hard optimization problems, which know great importance these past few decades in economy, management, manufacturing and many other fields. This algorithm gives a significant improvement to the basic genetic algorithm of J. Holland in order to imitate and simulate as close as possible the naturel selection phenomenal established in the theory of C. Darwin. Thus, in the evolution process of generations, the population should not keep a fixed size, but it should evolve over the generations. In the other hand, the population should contain several breeds of the species under study. Therefore, much kind of crossovers could be applied randomly such as crossover of pure or hybrid breeds. In addition, many types of mutation would be possible such as substitution, addition or deletion which could also happen randomly in the nature. The main idea is based on the maximal projection of the evolution theory on the optimization field to tackle complex problems. We aim to design flexible genetic algorithm by looking empirically for good compromise of adjusting the genetic parameters on sample cases.