Niching in evolutionary algorithms

Abstract

Niching techniques are the extension of standard evolutionary algorithms (EAs) to multimodal domains, in scenarios where the location of multiple optima is targeted and where EAs tend to lose population diversity and converge to a solitary basin of attraction. The development and investigation of EA niching methods have been carried out for several decades, primarily within the branches of genetic algorithms (GAs) and evolution strategies (ES). This research yielded altogether a long list of algorithmic approaches, some of which are bioinspired by various concepts of organic speciation and ecological niches, while others are more computational-oriented. This chapter will lay the theoretical foundations for niching, from the perspectives of biology as well as optimization, provide a summary of the main contemporary niching techniques within EAs, and discuss the topic of experimental methodology for niching techniques. This will be accompanied by the discussion of specific case-studies, including the employment of the popular covariance matrix adaptation ES within a niching framework, the application to real-world problems, and the treatment of the so-called niche radius problem.