On the performance of evolutionary algorithms with life-time adaptation in dynamic fitness landscapes

This work demonstrates how the efficiency of evolutionary algorithms in dynamic environments can be improved by use of life-time adaptation. Our results contradict the hypothesis that there would be a tradeoff between designing and tuning EAs for static and dynamic environments, in which improved efficiency in one type of environment would decrease the efficiency in the other. In contrast, we show that the inclusion of life-time adaptation can result in EAs that outperform traditional EAs in both static and dynamic environments. Since the performance of EAs with life-time adaptation in dynamic environments are currently poorly understood at best, we conduct an extensive evaluation of the performance of these EAs on combinatorial and continuous dynamic global optimization problems with well-defined characteristics. In doing so, we propose improved benchmark dynamic fitness functions for both the combinatorial and continuous domains, which we have termed random dynamics NK-landscapes and structured moving peaks landscapes, respectively.