Exploring Heterogeneity in Constrained Optimization: An Adaptive Two-Stage Surrogate-Assisted Evolutionary Algorithm
Chenyan Gu, Handing Wang · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2025
When tackling expensive constrained optimization problems (ECOPs), existing surrogate-assisted evolutionary algorithms (SAEAs) often overlook the differences in evaluation time between constraints and the objective. However, real-world optimization problems exhibit significant differences in evaluation time, termed heterogeneous expensive constrained optimization problems (HE-COPs). To explore the impact of this heterogeneity, we propose an adaptive two-stage SAEA based on differential evolution (ATS-SADE) to tackle scenarios where inequality constraints are relatively inexpensive to evaluate. In the proposed algorithm, by leveraging ample constraint evaluations, an adaptive surrogate-assisted repair strategy is developed to guide the population toward the feasible region in the first stage. Moreover, an adaptive two-stage switch strategy is introduced to transition to the second stage under opportune conditions, thereby adapting to different problems. Experimental results on benchmark problems from CEC 2010 and CEC 2017 show that ATS-SADE outperforms four state-of-the-art algorithms.