Utilizing surrogate-based low-fidelity models can significantly decrease the overall computational time required for optimizing problems where evaluating objectives and constraints is time-consuming. Traditionally, surrogate-assisted evolutionary multi- and many-objective optimization algorithms evaluate each new population member with either high- or low-fidelity models across all objectives and constraints. Recent research, however, indicates that mixed-fidelity evaluation, where certain objectives and constraints are assessed with high-fidelity models and others with low-fidelity models can lead to greater efficiency. This improvement arises because time saved from skipping too computationally cheap or expensive evaluations of constraints for largely feasible and infeasible solutions and objectives for infeasible solutions, can instead be allocated to assessing promising candidates near constraint boundaries and in non-dominated feasible regions. In this study, we introduce a mixed-fidelity selection metric that quantifies the potential advantages of evaluating each objective and constraint for each population member individually. This metric incorporates the likelihood of a solution dominating its neighbors, computational cost, surrogate model error, and the extent of constraint violation. We validate our approach with results on test and engineering design problems ranging from two to 10 variables, with up to eight objectives and 10 constraints. We compare our method against state-of-the-art surrogate-assisted evolutionary algorithms. The findings suggest that this approach offers a promising new direction for surrogate-assisted evolutionary multi- and many-objective optimization research.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号