Transfer Search Directions Among Decomposed Subtasks for Evolutionary Multitasking in Multiobjective Optimization
Yanchi Li, Wenyin Gong, Qiong Gu · Proceedings of the Genetic and Evolutionary Computation Conference · 2024
Evolutionary multitasking has attracted much attention over the past years due to its inter-task knowledge transfer capability. In this area, multiobjective multitask optimization (MO-MTO), aims to handle multiple multiobjective optimization tasks faster and better simultaneously via population synergies among tasks. Existing multiobjective multitask evolutionary algorithms (MO-MTEAs) for MO-MTO mostly transfer positions, i.e., decision variables, which may invoke negative knowledge transfer on tasks with low optimal domain similarities. However, such low similarities are common in practice. To address this issue, this paper proposes a new MO-MTEA, named MTEA/D-TSD, which transfers search directions, rather than positions, among decomposed subtasks for MO-MTO. In addition to position-neighborhood in the decomposition method, MTEA/D-TSD constructs and adaptively updates the search-direction-neighborhood for each subtask. It transfers successful search directions among neighbor subtasks to accelerate population evolution. Moreover, to further improve the efficiency of knowledge transfer, a transfer rate self-adaptation strategy is designed for MTEA/D-TSD. Experimental results on MO-MTO benchmark problems and a real-world application of sensor coverage problems demonstrated the superior performance of MTEA/D-TSD against state-of-the-art MO-MTEAs.