关键词: 约束多目标优化, Kriging模型, Pareto前沿, 全局优化, 期望超体积改进, 可行性概率

Abstract: Aiming at the multi-objective simulation optimization problem of complex engineering systems,a global multi-objective optimization algorithm based on Kriging model was proposed by combining optimization process with trial process.In this algorithm,the proposed infill sampling criteria was used to sequentially add new trials to handle constraints and to approximate true Pareto sets,which could help the algorithm find high-quality Pareto sets in very limited trials.By considering the feasibility probability,the spacing distances and the Kriging model's prediction uncertainty of trial points,one infill sampling strategy was designed to explore the disconnected feasible regions effectively.An infill sampling criterion was proposed by taking maximum expected hyper-volume improvement and feasibility probability as objectives,so as to balance the improvement of Pareto sets with confirming the boundaries of feasible regions.The proposed algorithm was tested on three typical benchmarks,and the effectiveness and efficiency were proved.

Key words: constrained multi-objective optimization, Kriging model, Pareto front, global optimization, expected hyper-volume improvement, probability of feasibility