Fuzzy logic-based diversity-controlled self-adaptive differential evolution

This article presents a novel method using a fuzzy system (FS) to control the population diversity during the various phases of evolution. A local search is applied at regular intervals on an individual selected at random to aid the population in convergence. This diversity control methodology is applied to vary the crossover rate of self-adaptive differential evolution (SaDE). Three variants of the SaDE algorithm are proposed: (1) diversity-controlled SaDE (DCSaDE); (2) SaDE with local search (SaDE-LS); and (3) diversity-controlled SaDE with local search (DCSaDE-LS). The performance of the proposed algorithms is analysed using a set of unconstrained benchmark functions with respect to average function evaluations, success rate and the mean of the objectives of 30 independent trials. The DCSaDE-LS algorithm had a better success rate for high-dimensional multimodal problems and conserved the number of function evaluations required for most of the problems. It is compared with other popular algorithms and the outcome of the proposed DCSaDE-LS algorithm is validated using non-parametric statistical tests. MATLAB codes for the proposed algorithms may be obtained on request.