Evolutionary optimization with hierarchical surrogates

Abstract The use of surrogate models provides an effective means for evolutionary algorithms (EAs) to reduce the number of fitness evaluations when handling computationally expensive problems. To build surrogate models, a modeling technique (e.g. ANN, SVM, RBF, etc.) needs to be decided first. Previous studies have shown that the choice of modeling technique can highly affect the performance of the surrogate model-assisted evolutionary search. However, one modeling technique might perform differently on different problem landscapes. Without any prior knowledge about the optimization problem to solve, it is very hard to decide which modeling technique to use. To address this issue, in this paper, we propose a novel modeling technique selection strategy in the framework of memetic algorithm (MA). The proposed strategy employs a hierarchical structure of surrogate models and can automatically choose a modeling technique from a pre-specified set of modeling techniques during the optimization process. A mathematic analysis is given to show the effectiveness of the proposed method. Moreover, experimental studies are conducted to compare the proposed method with two other modeling technique selection methods as well as three state-of-the-art optimization algorithms. Experimental results on the used benchmark test functions demonstrate the superiority of the proposed method.