MSSRGO: A multimeta-model-based global optimization method using a selection-rank-based infill sampling strategy

In recent years, multimeta-model-based global optimization (MMGO) methods have received increasing attention due to their good performance in solving expensive black box problems. In this article, a multimeta-model-based global optimization method using the selection-rank-based infill sampling strategy (MSSRGO) algorithm is proposed to obtain more precise solutions with satisfactory computing costs for expensive black box problems. The MSSRGO utilizes three basic meta models (kriging (KRG), radial basis function (RBF) and polynomial response surface (PRS)) to capture the complex relationship between design variables and objective functions. In each iteration, first, two reduced subspaces are identified and used alternately with the whole design space. Then, a novel selection-rank-based sampling strategy and a greedy searching strategy in promising regions are proposed to obtain new potential points. Moreover, fifteen typical benchmark functions (six low-dimensional problems and nine high-dimensional problems) are employed to test the performance of MSSRGO. Compared with one well-known (EGO) and four recent (MSEGO, MGOSIC, MDSD and EMMGO) algorithms, the numerical experiments and engineering application show that the MSSRGO has superior searching precision, the same or lower computing costs and strong robustness.