適応DEにおける確率的なパラメータ調整法とその評価

抄録

Differential Evolution (DE) is a population-based stochastic search method for real-valued function optimization. Like other metaheuristic algorithms, DE finds optimal or near-optimal solutions without a priori knowledge about the function being optimized. However, DE generally shows largely different performance according to the DE parameters adopted. Therefore, various DE variants have been developed in order to obtain more stable and better performance. A DE variant called SHADE is one of the highly competitive DE variants so far. SHADE introduces parameter archives for parameter adaptation to generate better optimization results. In this paper, SHADE is extended in such a way that parameter archives are managed by novel three strategies so that DE parameters are robust against fixation which may occur by trapping the evolutionary search into local optima. We call this method the robust SHADE, i.e., RSHADE. The computer simulations are conducted to examine the performance of RSHADE on 28 benchmarks.