Multi-objective Multi-Task Particle Swarm Optimization Based on Objective Space Division and Adaptive Transfer

Evolutionary multi-task optimization (EMTO) solves multiple related optimization problems/tasks in an evolutionary algorithm framework simultaneously with knowledge transfer. The effectiveness of knowledge transfer is a determining factor in the success of EMTO. Designing proper knowledge transfer strategy is a big challenge especially for multi-objective optimization problems, where the intensity, timing, and source selection of knowledge transfer must be taken good care of. To address these issues, this study introduces MOMTPSO, an innovative algorithm that integrates objective space division and adaptive transfer within a multi-objective multi-task optimization paradigm. The self-evolution efficiency of the optimization tasks determines the intensity and timing of knowledge transfer in MOMTPSO. Moreover, based on the particle distributions in the objective space, MOMTPSO selects guiding particles of low density as the transferred knowledge in the subspace to improve the swarm diversity and convergence. By considering the distance between tasks, MOMTPSO incorporates adaptive tuning of the acceleration coefficient of the guiding particle to facilitate effective knowledge transfer, aiding the target task in escaping local optima. Comparisons are made between MOMTPSO and other cutting-edge multi-objective EMTO algorithms on four multi-objective multi-task optimization benchmarks. The obtained results substantiate the effectiveness of MOMTPSO. The source code for MOMTPSO can be accessed at https://github.com/CIA-SZU/YJB.