Abstract: Evolutionary multi-task optimization has emerged as one of the research hotspots in the field of computational intelligence in recent years,with its principle being to enhance the efficiency of algorithms in simultaneously solving multiple tasks through knowledge transfer between tasks.Since improper selection of transfer knowledge can reduce positive knowledge transfer between tasks,how to appropriately select transfer knowledge has become a key research direction.Additionally,during the algorithm’s evolutionary process,single-layer population reduction is insufficient to sustain the algorithm’s efficient optimization performance over the long term.Based on this,this paper proposes an evolutionary multi-task optimization algorithm(MTDE-MCT) based on transfer knowledge selection and population reduction.Firstly,the task population is initialized,and fitness evaluation is conducted,utilizing a combined index based on Manhattan distance and fitness values for the selection of transfer knowledge.Next,a subpopulation alignment strategy is applied to eliminate feature differences in transfer individuals between tasks.Finally,a multi-layer population reduction strategy is proposed,which linearly reduces the task population size based on the algorithm’s evolutionary stage.To validate the performance of the proposed algorithm,comparisons are made with classic algorithms from recent years using the CEC2017 and WCCI2020 problem test sets.The experimental results demonstrate that the proposed algorithm exhibits strong competitiveness in solving multi-task optimization problems.

Key words: Evolutionary algorithm, Multi-task optimization, Transfer knowledge selection, Subpopulation alignment, Multi-layer population reduction