Dynamic niching in evolution strategies with covariance matrix adaptation

Evolutionary algorithms (EAs) have the tendency to converge quickly into a single solution in the search space. However, many complex search problems require the identification and maintenance of multiple solutions. Niching methods are the extension of EAs to address this issue. In our study, we propose an evolution strategy (ES) niching method, based on the covariance matrix adaptation (CMA) mechanism. We analyze our algorithm, introduce an experimental setup, and compare its performance with a previous ES niching method, known as the ES dynamic niching algorithm. In our comparison we introduce for the first time a new analytical tool for niching analysis, and in particular the early niching formation process. Based on successful data fit, we propose the well-known logistic model to describe our experimental results.