A hybrid criterion-based sample infilling strategy for surrogate-assisted multi-objective optimization

High-quality Pareto front is always pursued when solving multi-objective optimization problems with surrogate-assisted multi-objective optimization (SMOO). For this purpose, a hybrid criterion-based sample infilling strategy is proposed to improve the predicted performance of surrogate model. Two infill criteria are integrated in proposed strategy, one is the expected improvement matrix-based infill criterion, by which a new sample point is generated through maximizing the expected improvement function of the non-dominated solutions in objective space. The other one is the Euclidean distance-based infill criterion, by which a newly different sample point is generated through perturbing the non-dominated solutions in design space. At each iteration in the SMOO process, two promising infilling sample points are obtained by the proposed strategy with the filter strategy, and the corresponding responses are further evaluated with the real simulation models, then the sample set and surrogate models will be updated sequentially. According to the results of the numerical validations, the SMOO algorithm with the proposed strategy has the competitive capabilities in terms of high-quality Pareto front and computational stability. Finally, the hybrid criterion-based sample infilling strategy for SMOO is applied to solve a high-dimension multi-objective optimization problem related to lightweight design of an electric bus body structure.