A Special Point-Guided Surrogate-Assisted Multi-Objective Evolutionary Algorithm for Complex Pareto Fronts

Abstract

In some real applications, multi-objective optimization problems are computationally expensive，called expensive multi-objective optimization problems(EMOPs). Although scholars have proposed many surrogate-assisted evolutionary algorithms to handle EMOPs, their ability to handle EMOPs with complex Pareto fronts is still not ideal,due to ignoring the influence of special points on Pareto front. In view of this, the paper studies a special point-guided surrogate-assisted multi-objective evolutionary algorithm. First, an adaptive integrated method of global surrogate model based on non-dominance ranking is developed to improve its prediction accuracy on key search regions. Second, two local search strategies driven by special points, i.e., the knee-point-driven exploitation strategy with confidence interval, and the discontinuous-point-driven exploration strategy with interval optimization, are proposed to improve the search efficiency of population on key regions. Moreover, a new adaptive setting method of reference vector is designed to improve the utilization efficiency of individuals. The proposed algorithm are compared with 7 existing algorithms on 20 benchmark functions and the building energy-efficient design problem, experimental results show that it can obtain highly-competitive Pareto optimal solutions with less computational costs.

Keywords: Expensive optimization, Multi-objective, Evolutionary algorithm, Surrogate model, Special point