An evolutionary multitasking optimization algorithm via reference-point based nondominated sorting approach

Multiobjective multifactorial evolutionary algorithm (MOMFEA), which solves multiple tasks simultaneously based on a single population, has received considerable attention in recent decades. However, the negative transmission usually leads to slower convergence or worse distribution. To make use of the potential similarity between different tasks, this paper proposes an enhanced version for the MOMFEA using a reference-point based nondominated sorting approach (denoted as MFEA-RP). By using Multiple Dimensional Scaling, subtasks in different dimensions can be optimized simultaneously with a single set of reference points. The efficiency of the method is substantiated by multiobjective benchmark problems and practical instances. In most of the test probability, MFEA-RP converges faster to the true Pareto front. Better-distributed solutions are successfully found, which indicates the better representativeness to the solution space.