A Two-Phase Kriging Surrogate-Assisted Evolutionary Algorithm for Constrained Dynamic Optimization Problems

Constrained dynamic optimization problems (CDOPs) are prevalent in industrial applications, where dynamic objective and constraints are typically represented by differential equations, resulting in high computational costs. Consequently, CDOPs can be considered as expensive optimization problems (EOPs). To address these challenges, surrogate-assisted evolutionary algorithms are commonly employed, utilizing surrogate models to replace costly function evaluations. In this paper, we propose a two-phase Kriging surrogate-assisted evolutionary algorithm to solve CDOPs. In Phase 1, when no feasible solution exists in the current sample set, a hybrid lower confidence bound criterion is introduced to guide the algorithm in quickly identifying a feasible solution. Once a feasible solution is identified, Phase 2 employs a lower confidence bound criterion based on information entropy to enhance the optimality of the solutions. Two industrial problems are used to test the algorithm's effectiveness compared with other 3 surrogate-assisted evolutionary algorithms and PSO. Experimental results demonstrate that, while maintaining optimality, the proposed algorithm significantly reduces computational costs.