A Differential Evolution Based Self-Adaptive Multi-Task Evolutionary Algorithm

This paper proposes a novel self-adaptive evolutionary multi-task optimization algorithm based on differential evolution (SMTDE) for solving multiple different optimization problems or tasks simultaneously. The algorithm arranges a specific population and three differential strategies for each task. Among the three strategies, one is the transfer strategy and the others are non-transfer strategies. The transfer strategy is mainly responsible for utilizing the information of other tasks, and the two non-transfer strategies are responsible for accelerating convergence and improving the diversity of intra-task, respectively. Based on strategies, a self-adaptive mechanism is proposed to adjust the selection probabilities of the three strategies to reduce the harm of negative transfer and balance the diversity and convergence within the population. The experiment is conducted on a single-objective multi-task test suite. The experiment results show that SMTDE can find better solutions with a higher convergence rate in comparison with several competitive evolutionary multi-task optimization algorithms.