Crowding distance based promising solution selection in surrogate assisted asynchronous multi-objective evolutionary algorithm

This paper proposes an efficient solution selection method in a surrogate-assisted asynchronous multi-objective evolutionary algorithm. Our previous research proposed a novel multi-objective evolutionary algorithm that integrates a surrogate evaluation model with asynchronous approach, named as AELMOEA/D. AELMOEA/D constructs a surrogate model with extreme learning machine (ELM) and generates a promising solution by MOEA/D with a constructed ELM model. A generated promising solution is selected in the order of the indexes of the weighted vector of MOEA/D, and is evaluated asynchronously. In contrast to the previous method, the proposed method considers degree of search progress of each weight vector and selects a promising solution in a region where the search progress is insufficient. To evaluate the degree of the search progress, this study employs crowding distance, which is basically used in NSGA-II. To investigate the effectiveness of the proposed method, we conduct the experiment on a multi-objective optimization benchmark problem. The experimental result revealed that the proposed method can accelerate the convergence speed of the optimization without deteriorating the performance compared with the previous method.