Evolutionary Optimization and Dynamic Fitness Landscapes

Abstract

Evolutionary algorithms are a promising option for solving dynamic optimization problems. These problems have fitness landscapes whose topological features change dynamically with the run-time of the evolutionary algorithm. In this chapter, we study these landscapes by analyzing and quantifying their properties using topological and dynamical landscape measures such as modality, ruggedness, information content, dynamic severity and two types of dynamic complexity measures, Lyapunov exponents and bred vector dimension. Here, our main focus is on dynamic fitness landscapes that exhibit spatio-temporal chaotic behavior. We further discuss evolutionary algorithms and modifications needed to make them fit to perform in dynamic landscapes and present numerical experiments showing the algorithms’ performances. These results allow us to link the landscape measures to the behavior of the evolutionary algorithms.