Multipopulation-based multi-tasking evolutionary algorithm

Multi-tasking optimization (MTO) has attracted more and more attention from researchers in the area of evolutionary computing. The main factor affecting the success of MTO is knowledge transfer. Nevertheless, knowledge transfer between tasks has positive and negative effects on tasks that are solved simultaneously. In multi-task evolutionary optimization, the negative migration can be suppressed to a certain extent by adjusting random mating probability between tasks, but the negative migration between tasks cannot be completely avoided. This paper proposes a new multi-population-based multi-task evolutionary algorithm (MPEMTO) to weaken the impact of negative knowledge transfer between tasks. The MPEMTO has a novel dual information transfer strategy, an adaptive knowledge screening mechanism, an extended adaptive mating strategy, and a computational resource allocation method. MPEMTO first applies adaptive mating strategy and dual information migration strategy to control the transfer of knowledge between tasks and then applies a transfer information screening mechanism to screen the transfer information to achieve effective use of the transfer information between tasks. The effectiveness of MPEMTO is compared with eight excellent algorithms on single-object MFO test problems. The experimental results demonstrate that the performance of the MPEMTO algorithm is very competitive on most optimization problems.