A dynamic multi-objective evolutionary algorithm based on two-stage dimensionality reduction and a region Gauss adaptation prediction strategy

Abstract

Dynamic multi-objective optimization evolutionary algorithms (DMOEAs) have attracted increasing attention due to their superior performance in real-world applications. It is challenging to rapidly track the changing Pareto set (PS) and Pareto front (PF) in DMOEAs. However, most existing studies only focus on the position change in the PS. To address this problem, we define a series of new PS region changes. Based on these definitions, we propose a dynamic multi-objective evolutionary algorithm based on two-stage dimensionality reduction and a region Gauss adaptation prediction strategy (DRPS). Specifically, we estimate the possible type of PS changes at the next moment by processing historical information. Afterward, the basic scaling population PSB is generated by dimensionality reduction, while parameters including the scaling factor and density change rate are calculated. Combining these parameters with a Gaussian distribution, the algorithm adaptively adjusts the sampling probability of individuals in PSB, achieving prediction with different types of PS changes. We conducted extensive experiments and compared classical algorithms, the latest algorithms, and algorithms with the same type in 25 test functions. The results demonstrate that the proposed algorithm outperforms the compared algorithms in most cases, while implying that the newly defined PS region change is reasonable.