Improvement of Genetic Algorithm Using a Fuzzy Control Combined with Coevolutionary Algorithm

In order to achieve the best performance in genetic algorithm, proper determination of parameters is necessary. This paper addresses the intelligent determination of the crossover probability between two selected parents in each generation. Unlike most existing techniques that utilize the diversity characteristics of each generation to determine the crossover probability throughout the current generation, this paper defines some novel phenotype and genotype features, and develops a zero-order Takagi-Sugeno fuzzy controller to derive the proper crossover probability for each selected parent pair. As such, each pair has a unique probability parameter that results in the flexibility of the standard genetic algorithm depending on the region being searched and avoids premature convergence. In addition, in the proposed method, the consequent part of the fuzzy rules is not fixed but is generated through a coevolutionary process and simultaneously with the decision variables of the optimization problem. This enhances the efficiency of the proposed method. The simulation results on a set of optimization benchmarks show the performance of this method. Its effectiveness is also investigated by applying it to the complicated problem of terrain avoidance/terrain following fly.