An adaptive stochastic search algorithm

Abstract

A new population-based stochastic search algorithm is developed which automatically adjusts search domains of individuals in terms of current search information and individual preferences in the search process. It achieves a proper balance between global exploration and local exploitation in a simple and natural way by adaptively varying the position and size of the neighborhood space of each individual and gradually shrinking to global optima. It allows individuals to randomly enlarge their search radiuses in the search process and to have more chances to jump out of the likely local optima when dealing with some difficult tasks. The test results on five classical benchmark functions demonstrate the excellent global optimization ability, high search efficiency and good stability of the algorithm. It performs significantly better than PSO, FS and GAFS. With the virtue of inherent robustness, implicit parallelism and easy implementation, the proposed algorithm is applicable to complicated high-dimensional multimodal optimization problems.