A dynamic selection strategy for classification based surrogate-assisted multi-objective evolutionary algorithms

Multi-objective problems (MOPs) consist of at least two objectives and they conflict with each other. A class of expensive problems is one of multi-objective problem class, that requires high costs for large calculations, large space, and large number of objectives. The useful and popular method for expensive problems is the use of surrogate models. There are many techniques used in surrogate models, such as RBF, PRS, Kriging, SVM, ANN..., that are used and achieve relatively good results. One problem, however, is the choice of time to update the model, the selection of reference data... Those actions are very important in an evolutionary process. The parameters for control those action usually predefined, so it might reduces the adaptability of algorithms. In this paper, we analyze a dynamic selection method of reference solutions based on information from the population, the evolution process to produce better results, with each different problem.