Analysis of Relation between Prediction Accuracy of Surrogate Model and Search Performance on Extreme Learning Machine Assisted MOEA/D

In recent years, evolutionary algorithms have been used for many real-world problems, but it takes enormous computation time to obtain the optimal solution due to its high calculation cost. Multi-objective evolutionary algorithms using surrogate models have been studied to reduce the computation time for the optimization. ELMOEA/D is one of the surrogate-assisted multi-objective evolutionary algorithms. ELMOEA/D combines MOEA/D with an extreme learning machine (ELM). This paper analyzes the relation between the estimation accuracy of the surrogate model and the search performance of ELMOEA/D. We experiment on several well-known multi-objective benchmark problems and compare the different number of generations. The experimental results reveal that the estimation accuracy and the search performance decrease as the number of generations increase.