A general selection method for mutation strategy in differential evolution

Abstract

How to balance exploration and exploitation is a key issue for evolution algorithm including differential algorithm (DE). Many researchers propose various improved mutation strategies to solve this issue for DE. Most of them can be classified as deterministic rules. That is to say, they select individuals according to predetermined methods and so the balance is static. However, different evolution stages require different balance between exploration and exploitation. In order to solve this problem, a general selection method named adaptive stochastic ranking based mutation strategies in DE(ASR-DE). In ASR-DE, it uses stochastic ranking method to rank all individuals according to their contribution in exploration and exploitation. The parameter P f in stochastic ranking is adaptive controlled by a transform version of success rate. The individuals with the smaller ranking are more likely to be selected. 28 functions of CEC2013 is used here to verify the validity of testing method. The test results show that ASR-DE improves the standard DE and improved DE comparing with other methods.