A genetic algorithm with neutral mutations for massively multimodal function optimization

The paper presents an extended genetic algorithm (GA) for massively multimodal function optimization. The proposed GA includes two features; one introduces redundancy into string representation, and the other divides the population into subpopulations only for the stage of selection and reproduction of each generation. The mechanism develops the behavior of finding deceptive hyperplanes and escaping from them using large genetic transitions to the complements to them in the population. The influence of genetic drift is avoided by adopting the elitist strategy in each subpopulation. An experiment is given for illustrating the efficiency of the proposed method for a massively multimodal problem