Author(s)

T. Maruyama & E. Kita

Abstract

This paper describes the extended Stochastic Schemata Exploiter (ESSE), which has the improved algorithm of the Stochastic Schemata Exploiter (SSE). The algorithm of the ESSE is composed of the original SSE and the ESSE operations. There are seven ESSE algorithms. In the previous study, the authors compared seven ESSE algorithms in some test problems. The ESSE-c1 algorithm shows the best search performance among them. In this paper, the ESSE-c1 algorithm is compared with BOA and SSE in order to confirm the features. 1 Introduction In most of the combinational optimization problems, the objective function spaces have so-called \“big valley structure” [1]. In the problems with big valley structure, there is often the real (global) optimum solution near quasi-optimal solutions. The evolutionary algorithms are considered to be effective for such optimization problems [2–4]. Stochastic Schemata Exploiter (SSE) has been presented by Aizawa in 1994 [5]. In the traditional SGA, better individuals are selected as parents from a population and genetic operators generate new individuals from them. The SSE has a different algorithm than the SGA. In the SSE, the sub-populations are determined according to the descending order of the fitness of the individuals. Common schemata are extracted from the sub-populations and new individuals are generated from them. Selection and crossover operations are not necessary for the SSE. Since the SSE algorithm tends to search better solution near good solutions which are already found, it is adequate for the problems with function spaces of big valley structure. The SSE has two attractive features. First, the control parameters are relatively small because selection and crossover operators are not necessary. Second, the

Keywords