A study on metamodeling techniques, ensembles, and multi-surrogates in evolutionary computation

摘要: Surrogate-Assisted Memetic Algorithm (SAMA) is a hybrid evolutionary algorithm, particularly memetic algorithm that employs surrogate models in the optimization search. Since most of objective function evaluations SAMA are approximated, search performance likely to be affected by characteristics used. In this paper, we study using different meta modeling techniques, ensembles, and multi-surrogates SAMA. particular, consider SAMA-TRF, model management framework incorporates trust region scheme for interleaving use exact with computationally cheap local during searches. Four metamodels, namely Gaussian Process (GP), Radial Basis Function (RBF), Polynomial Regression (PR), Extreme Learning Machine (ELM) neural network used study. Empirical results obtained show while some metamodeling techniques perform best on particular benchmark problems, ensemble metamodels multisurrogates yield robust improved solution quality problems general, same computational budget.