Evolutionary Optimization of Dynamic Multiobjective Functions

Abstract

Many real-world problems show both multiobjective as well as dynamic characteristics. In order to use multiobjective evolutionary optimization algorithms (MOEA) efficiently, a systematic analysis of the behavior of these algorithms in dynamic environments is necessary. Dynamic fitness functions can be classified into problems with moving Pareto fronts and Pareto sets having varying speed, shape, and structure. The influence of the dimensions of the objective and decision space is considered. The analysis will focus on standard benchmark functions and newly designed test functions. Convergence and solution distribution features of modern MOEA, namely NSGA-II, SPEA 2 and MSOPS using different variation operators (SBX and Differential Evolution), will be characterized using Pareto front metrics. A new path integral metric is introduced. Especially the ability of the algorithms to use historically evolved population properties will be discussed.