State information-driven surrogate-assisted differential evolution for computationally expensive constrained optimization problems

Abstract

In this paper, a state information-driven surrogate-assisted differential evolution called SI-SADE is proposed for solving expensive constrained optimization problems, in which both the population state and adaptive search mechanism are respectively evaluated and designed based on the feasibility and state information. Firstly, the multiple subpopulations are obtained by comprehensively considering the three different population states, i.e., infeasible, partially feasible, and fully feasible, and the diversified indicators of population individuals. Secondly, different ensemble mutation and environmental selection operations are tailored specially for subpopulations where both an inner evolution-driven parent expansion and update rate-based surrogate switch strategies are designed to regulate the search ability of the algorithm. Furthermore, to bypass the hard obstacles caused by complex constraints, a pure objective-based search rectification is used to locate the possible feasible region in the direction of minimizing objective value. Therefore, the SI-SADE achieves an adaptive balance between feasibility and convergence. Systematic experimental results on both the IEEE CEC2010 and CEC2017 benchmark problems demonstrate the high competitiveness of SI-SADE. More importantly, the SI-SADE performs excellently in solving a real-world case.