A Multiobjective Multitask Evolutionary Algorithm Based on Decomposition and Multivariate Gaussian Distribution

Evolutionary multiobjective multitask optimization (EMTO) has attracted widespread attention in recent years, which solves multiple tasks simultaneously in a single population. How to extract effective knowledge and recognize valuable transferred solutions is the key to enhance the performance of EMTO. However, few research studies consider these two issues at the same time. To fill this research gap, we propose a novel multiobjective multitask evolutionary algorithm based on decomposition and multivariate Gaussian distribution, called MTEA-DMG, in which a candidate transferred solution set is generated using resource allocation according to the priority of subproblems to multiple tasks. Then, one most valuable knowledge transfer carrier is selected by online statistical estimation of distribution density. The experimental results on a set of benchmark problems with different degrees of similarity show that MTEA-DMG is superior to other state-of-the-art EMTO algorithms.