A surrogate-assisted evolutionary algorithm with Gaussian process regression and diversity search for large-scale expensive optimization

Expensive optimization problems (EOPs) need time-consuming simulations or expensive physical experiments to evaluate candidate solutions, posing a challenge for optimization. Surrogate-assisted evolutionary algorithms (SAEAs) have shown desirable performance in solving EOPs. However, most existing SAEAs are initially designed for low-dimensional EOPs and rarely consider handling large-scale EOPs. To fill this gap, this work proposes an ensemble surrogate-assisted evolutionary algorithm with a diversity search (DS-SAEA) and a lower-confidence-bound (LCB)-based Gaussian process regression (GPR) for large-scale EOPs. To handle large-scale complex optimization, the proposed DS-SAEA combines a surrogate-based global search and a surrogate-based local search taking into account both exploration and exploitation. On one hand, the proposed surrogate-based global search combines LCB-based GPR and radial basis function (RBF)-based ensemble surrogate to realize a more accurate and effective approximation to the entire landscape of EOPs. On the other hand, the local RBF surrogate-based search is used to search the local region finely. The RBF-based ensemble surrogate is composed of three different RBF surrogates with different features to make the search more diverse and robust. Further, an improved JADE is proposed with a more uniform initial population to improve the search capability. Experimental results on many well-known benchmark problems have shown the superior performance of the proposed algorithm over seven state-of-the-art SAEAs on most problems.