Heterogeneous Multitask Optimization Scheduling Based on Adaptive Knowledge Transfer

Multitask optimization (MTO) algorithm can realize the parallel processing of multiple optimization tasks by mining the relationship between each task. However, multiple tasks in practical applications usually have heterogeneous characteristics, and it is difficult for existing multitask optimization algorithms to effectively analyze and utilize the relationship between these heterogeneous tasks. Therefore, a heterogeneous multitask optimization scheduling algorithm based on adaptive knowledge transfer (HMTO-AKT) is designed in this paper. First, cross-task mutual information is proposed to assess the importance of dimensions. By comparing the mutual information of different dimension combinations, the most important dimension information of each dimension for the target task is obtained to deal with the decision space heterogeneous. Second, an adaptive knowledge transfer strategy is proposed to promote positive transfer. By evaluating the knowledge validity and the evolution state of the population, the effective knowledge transfer is promoted to solve target space heterogeneous. Third, a new particle encoder mode is designed. The operation of HMTO-AKT on scheduling problem is realized through encoder. Finally, a heterogeneous MTO-VRP is constructed based on the Solomon benchmark test set, and comparative tests are carried out on this problem. The results show that HMTO-AKT has great advantages in solving heterogeneous MTO problems.