A Multitasking-Based Constrained Multi-Objective Evolutionary Algorithm With Forward and Backward Stages
Xiaoyu Zhong, Xiangjuan Yao, Kangjia Qiao, Dunwei Gong
Abstract
Solving constrained multi-objective optimization problems (CMOPs) via evolutionary multitasking optimization (EMTO) algorithm is a meaningful attempt due to the fact that knowledge learning and transfer across the related optimization exercises can improve the efficiency of others. However, the utilization of EMTO in CMOPs is still in its infancy. To fill this research gap, an EMTO-based constrained multi-objective optimization framework including forward and backward stages (CEMTFB) is developed to address an original CMOP task together with two helper CMOP tasks. Firstly, the original task and a helper task evolve forward simultaneously with and without constraints, responsible for exploring well-converged and well-distributed feasible and infeasible solutions, respectively. Then, in the backward stage, a novel reverse haulage strategy is designed for another helper task to conduct a search within the promising areas that are not dominated by any examined feasible solution, thereby collaborating with the original task to approach the constrained Pareto front from two complementary directions. Moreover, a dynamic knowledge transfer strategy is proposed to coordinate the interaction between the original and helper tasks. Finally, comprehensive experiments are conducted on 45 benchmark functions and 19 real-world CMOPs, and the comparison against seven state-of-the-art peer methods demonstrates the superior or at least competitive performance of CEMTFB.