A Special Points-Based Hybrid Prediction Strategy for Dynamic Multi-Objective Optimization

Abstract

Dynamic multi-objective optimization problem (DMOP) is such a type of optimization problems that multiple contradictory objectives change over time. This paper designs a special point-based hybrid prediction strategy (SHPS) integrated into the decomposition-based multi-objective optimization algorithm with differential evolution (MOEA/D-DE) to handle DMOPs, which is denoted as MOEA/D-DE-SHPS. In the SHPS, when historical information is insufficient to establish prediction model of population prediction strategy (PPS), the prediction (PRE) and variation (VAR) method are adapted to generate the initial population of the new environment. Meanwhile, the PPS predicts the whole population of new environment according to the history information collected from past environments; therefore, once collected historical information is inaccurate, the predicted population may be located in the wrong search region. To overcome the shortcoming, we propose the special point-based strategy in which the initial population of the new environment consists of two parts of individuals: the predicted special points and the predicted population by PPS (except the special points). The empirical results show that MOEA/D-DE-SHPS is promising for handling DMOPs.