Evolutionary Multitasking with Scale-Adaptive Subspace for Large-Scale Optimization

Abstract

Large-scale optimization is an important research topic in the field of evolutionary computation which has been widely applied in many scientific and engineering fields. In large-scale optimization, many research studies on simplifying the search space for large-scale optimization problems (LSOPs) have been proposed, such as cooperative co-evolution, and dimension reduction. However, the validity of simplified solution spaces with different scales in the evolutionary process is not fully explored to guide the search direction. Therefore, this paper puts forward an evolutionary multitasking method with scale-adaptive subspace (EMT-SAS) for LSOPs, which uses a set of subspaces with different accuracy scales to cooperatively assist original problem search. The basic idea is to adopt the low-accuracy scale subspace to fast locate promising regions and utilize the high-accuracy scale subspace to refine the solution accuracy. Moreover, an archive management strategy is also proposed to select training samples, which can further enhance the diversity of the algorithm. The performance of the proposed EMT-SAS is evaluated on the CEC2013 large-scale benchmark problems. The experimental results have demonstrated the superiority of EMT-SAS when compared with other advanced algorithms.