Dichotomy Guided Based Parameter Adaptation for Differential Evolution

Abstract

Differential evolution (DE) is an efficient and powerful population-based stochastic evolutionary algorithm, which evolves according to the differential between individuals. The success of DE in obtaining the optima of a specific problem depends greatly on the choice of mutation strategies and control parameter values. Good parameters lead the individuals towards optima successfully. The increasing of the success rate (the ratio of entering the next generation successfully) of population can speed up the searching. Adaptive DE incorporates success-history or population-state based parameter adaptation. However, sometimes poor parameters may improve individual with small probability and are regarded as successful parameters. The poor parameters may mislead the parameter control. So, in this paper, we propose a novel approach to distinguish between good and poor parameters in successful parameters. In order to speed up the convergence of algorithm and find more "good" parameters, we propose a dichotomy adaptive DE (DADE), in which the successful parameters are divided into two parts and only the part with higher success rate is used for parameter adaptation control. Simulation results show that DADE is competitive to other classic or adaptive DE algorithms on a set of benchmark problem and IEEE CEC 2014 test suite.