Study on Improvement of Memory Cell Control in Hybridization of Immune Algorithm and Gradient Search for Multiple Solution Search

Abstract

Many evolutionary computation methods have been proposed and applied to real world problems. But gradient methods are still effective in problems involving real-coded parameters. In addition, it is desirable to find not only an optimal solution but also plural optimal and semi-optimal solutions in most real world problems. Although a hybrid algorithm combining Immune Algorithm (IA) and Quasi-Newton method (QN) has been proposed for multiple solution search, its memory cell control sometimes fails to keep semi-optimal solutions whose evaluation value is not so high. In addition, because the hybrid algorithm applies QN only to memory cell candidates, QN can be used as local search operator only after global search by IA. This paper proposes an improved memory cell control which restricts existence of redundant memory cells, and a QN application method which uses QN even in early search stage. Experimental results have shown that the hybrid algorithm involving the proposed improvements can find optimal and semi-optimal solutions with high accuracy and efficiency even in high-dimensional multimodal functions involving epistasis.