Evolutionary algorithms with clustering for dynamic fitness landscapes

Abstract

Interest on dynamic multimodal functions risen over the last years since many real problems have this feature. On these problems, the goal is no longer to find the global optimal, but to track their progression through the space as closely as possible. This paper presents three evolutionary algorithms for dynamic fitness landscapes. In order to mantain diversity in the population they use two clustering techniques and a macromutation operator. Besides, this paper compares two crossover operators: arithmetic and multiparents two points, respectively. Effectiveness and limitations of each algorithm are discuss and analyzed.