A bagging-based surrogate-assisted evolutionary algorithm for expensive multi-objective optimization

It is a big challenge for multi-objective evolutionary algorithms to solve expensive multi-objective optimization due to high computational cost. To effectively address expensive multi-objective optimization, this work proposes a novel surrogate-assisted evolutionary algorithm (SAEA), named bagging-based SAEA (B-SAEA). In the proposed method, bagging is introduced to construct high-quality surrogate ensembles for each expensive objective under a limited number of training points. Thereafter, an evolutionary search is applied to fully search for the constructed surrogate ensembles with the help of generation-based search strategy. Thus, surrogate ensembles and evolutionary search can be seamlessly integrated. In addition, a niche-based infill solutions selection strategy is proposed to select the promising points as the infill solutions for real fitness evaluations. As a result, a good balance between convergence and diversity can be achieved within a limited computational budget. Experimental results on commonly used benchmark test problems and real-world engineering application have demonstrated that the proposed method performs competitively compared with other state-of-the-art methods.