Adaptive differential evolution with hybrid rules of perturbation for dynamic optimization

Abstract

This work presents a differential evolution (DE) algorithm equipped with a new perturbation operator applied for dynamic optimization. The selected version of DE, namely the jDE algorithm has been extended by a new type of mutation mechanism which employs random variates controlled by the á-stable distribution. Precisely, in the modified version of jDE the population of individuals consist of two types of members: a small number of those which undergo the new mutation procedure and much larger number of the remaining ones which are mutated according to the regular DE mutation. This hybrid structure of population makes the algorithm more effective for some types of the dynamic environments. The experiments were performed for two well known benchmarks: Generalized Dynamic Benchmark Generator (GDBG) and Moving Peaks Benchmark (MPB) reimplemented together as a new benchmark suite Syringa. Obtained results show advantages and disadvantages of the new approach. © Springer-Verlag Berlin Heidelberg 2013.