Fuzzy System Multitask Optimization Based on Mixture Strategy

Evolutionary multitasking optimization (EMTO) received considerable attention for its efficient problem-solving ability, but the most important factor of EMTO is knowledge transfer (KT). Although current research has made some progress, it faces two challenges, first how to transfer shared knowledge in an efficient way to minimize negative effect. Second how to prevent premature convergence and maintain population diversity throughout the evolutionary process. To address these two challenges, this paper proposes a new multitask optimization based on mixture strategy (MTMS), which includes fuzzy system based on genetic algorithms (FSGA) and fuzzy system based on differential evolution (FSDE). The proposed algorithm combines the prominent local exploration ability of FSGA, and at the same time utilizes the differential vectors of FSDE to jump out of the local optimal region to explore the broad search space. They both have strategies for learning randomly from other dimensions, which makes them more effective in optimizing dimensionally misaligned problems. The experimental results on CEC2017 benchmark functions show that MTMS is significantly better than some powerful and classic EMTO algorithms proving the effectiveness of MTMS.