A Surrogate-Assisted Multi-Objective Evolutionary Algorithm with Multiple Reference Points

Some expensive multi-objective optimization problems (EMOPs) in real life have discontinuous Pareto Front (PF). However, due to the lack of specific strategies targeting the discontinuous parts of PFs, most existing evolutionary algorithms cannot solve this problem very well. To address this issue, this paper proposes a surrogate-assisted multi-objective evolutionary algorithm based on multiple reference points. Firstly, an identification strategy for the boundary points of discontinuous PFs is given to obtain spatial distribution characteristics of PFs. Subsequently, a two-phase surrogate model management strategy assisted by multiple reference points is proposed to select infilling samples with different characteristics for the global surrogate model at different stages of the algorithm, thereby helping the algorithm allocate limited computing resources reasonably. Additionally, a local exploitation strategy for discontinuous regions of PF is developed to enhance the accuracy of the algorithm. Finally, the proposed algorithm is used in 20 benchmark functions, and experimental results show that it can obtain a set of high-quality Pareto optimal solutions at a less computational cost.