A Proposal of Memory and Prediction Based Genetic Algorithm Using Speciation in Dynamic Multimodal Function Optimization

Abstract

It is a difficult problem for evolutionary algorithms to search an optimal solution in multimodal functions with dynamic environments, where individuals searchmore than one optimum and their fitness values change over time. In this paper, we propose a method of Memory and Prediction Based Genetic Algorithm Using Speciation. This method is extended with a case based memory and a meta learner for precise prediction of environmental change. Especially, the individuals in a memory consist of 4 kinds of predictors and they can adjust to the change of dynamic environment adaptively. Speciation has shown to be an effective technique for multimodal optimization. A niching method based on speciation can be used to classify a population into groups according to their similarity measured by a distance. In this paper, each group by speciation has a memory and the individuals stored in the memory can respond to the situation according to the dynamic environment. In order to verify the effectiveness, the method is examined to search for an optimal solutions in multimodal functions.