Adaptive Approaches Towards Better GA Performance in Dynamic Fitness Landscapes

Abstract

We review different techniques for improving GA performance. By analysing the fitness landscape, a correlation measure between parents and offspring can be provided, and we can estimate effectively which genetic operator to use in the GA for a given fitness landscape. The response to selection equation further tells us how well the GA will do, and combining the two approaches gives us a powerful tool to automatically ensure the selection of the right parameter settings for a given problem. In dynamic environments the fitness landscape changes over time, and the evolved systems should be able to adapt to such changes. By introducing evolvable mutation rates and evolvable fitness formulae, we obtain such systems. The systems are shown to be able to adapt to both internal and external constraints and changes.