Surrogate-Assisted Particle Swarm Optimization with Dual-Subspace Search for Large-Scale Expensive Optimization

Several meta-heuristic algorithms have gained attention for addressing large-scale optimization problems (LSOPs). However, their effectiveness is limited when dealing with LSOPs that involve expensive objective functions because the significant number of objective evaluations required before identifying the optimal solution. In this paper, two sets of subproblems are op-timized, assisted by surrogate models, to collaboratively explore the optimal solution space for large-scale expensive optimization problems (LSEOPs). A variant of particle swarm optimization (PSO), engineered to accelerate convergence, is proposed to search for the optimal solution for each subproblem. Moreover, a novel infill criterion is introduced, where the solution with the maximum expected improvement in the first set of subproblems and the solution with the minimum objective value discovered so far in the second set of subproblems are selected. Then, their mean position will be evaluated using the real objective function. Experimental findings on 15 CEC2013 benchmark problems show that our proposed approach outperforms four state-of-the-art algorithms in solving expensive large-scale optimization problems with up to 1000 decision variables.