A hybrid of genetic transform and hyper-rectangle search strategies for evolutionary multi-tasking

Abstract

Recently, evolutionary multi-tasking (EMT) has surfaced as a new search paradigm in the field of evolutionary computation to solve two or more tasks simultaneously. EMT algorithms accelerate the convergence of multiple optimization tasks by sharing useful knowledge among tasks, i.e., knowledge transfer is the key to the success of EMT algorithms. However, as the evolutionary search proceeds, the learnability of one task to others might decrease and the knowledge transfer becomes less efficient. To address this issue, this paper proposes a novel multi-factorial evolutionary algorithm by hybridizing two complementary strategies, namely genetic transform strategy and hyper-rectangle search strategy (MFEA-GHS). The proposed genetic transform strategy is applied in individual reproduction and aims to strengthen the knowledge transfer efficiency. Particularly, if two parent individuals are specific to different tasks, one parent individual is transformed to fit the other task by task space mapping. As such, higher-quality offspring individuals towards a target task can be generated with the transformed individual. The hyper-rectangle search strategy based on opposition learning is designed to perform efficient exploration and exploitation in both the unified search space and the sub-space of each task, which enables the population to search more unexplored regions. Comprehensive experiments are carried out on both single- and multi-objective EMT optimization problems. The experimental results demonstrate the efficiency of MFEA-GHS and the two proposed strategies.