A grey prediction-based evolutionary algorithm for dynamic multiobjective optimization

Abstract

Abstract Dynamic multiobjective optimization problems (DMOPs) usually involve multiple conflicting objectives that change over time. A good evolutionary algorithm should be able to quickly track the moving Pareto optimal front (POF) and Pareto optimal set (POS) over time. To solve DMOPs, a predictive method is proposed herein based on grey prediction model, which is composed of three essential ingredients. The first one is that the population is divided into multiple clusters, which can help the population to preserve diversity throughout the evolutionary process. The second one is that the individuals used to detect environmental changes are taken from different clusters, which in turn help the proposed algorithm to detect environmental changes more promptly and accurately. The third one is to build the grey prediction model by using the centroid point of each cluster when detecting the environmental change, and then generate the initial population. Empirical results show that the proposed algorithm can deal with dynamic environments and track the varying POS and POF effectively and efficiently, and achieve better performances on most test problems than several selected state-of-the-art algorithms.