Probability Matching-based Adaptive Strategy Selection vs. Uniform Strategy Selection within Differential Evolution

摘要

Different strategies can be used for the generation of new candidate solutions on the Differential Evolution algorithm. However, the definition of which of them should be applied to the problem at hand is not trivial, besides being a sensitive choice with relation to the algorithm performance. In this paper, we use the BBOB-2010 noiseless benchmarking suite to further empirically validate the Probability Matching-based Adaptive Strategy Selection {PM-AdapSS-DE) [4], a method proposed to automatically select the mutation strategy to be applied, based on the relative fitness improvements recently achieved by the application of each of the available strategies on the current optimization process. It is compared with what would be a timeless (naive) choice, the uniform strategy selection within the same sub-set of strategies.