Evolutionary Constrained Mult-Objective Optimization Based on Competitive Multitasking and Decomposition-Dominance
XU Jinyu, Hui Wang, Shitao Liao, Hangyu Liu, Yun Wang, Xinyu Zhou · 2024
During the past two decades, numerous constrained multi-objective evolutionary algorithms (CMOEAs) have been proposed for constrained multi-objective problems (CMOPs). However, most of the existing algorithms perform poorly when facing complex CMOPs. Inspired by evolutionary multitasking (EMT) and competitive multitaskin$\mathbf{g}$(CMT), a novel constrained multi-objective optimization algorithm based on CMT and decomposition-domination (CMT-MOEADD) is developed in this paper. Firstly, in the decomposition stage, two subtasks are designed based on the CMT framework, and a reward mechanism is designed to determine which task is selected to become the main task. One of the sub-tasks uses fuzzy constraint handling techniques to balance the relationship between objectives and constraints, while the other sub-task disregards constraints to find the unconstrained Pareto frontier (UPF). Knowledge transfer between tasks is then relied upon to guide the external profile towards the CPF. Finally, the set of populations and external archives in the dominance stage using the ε-MO based constraint handling technique evolve in the EMT framework to accelerate population convergence. In addition, we conduct an experimental study on two test suites from recent years to compare CMT-MOEADD with the state-of-the-art five algorithms. Experimental results demonstrate that our proposed CMT-MOEADD has superior or competitive performance.