A constrained multi-objective evolutionary algorithm based on decomposition and dynamic constraint-handling mechanism

Abstract

Constrained multi-objective optimization problems (CMOPs) are common in real-world engineering application, and are difficult to solve because of the conflicting nature of the objectives and many constraints. Some constrained multi-objective evolutionary algorithms (CMOEAs) have been developed for CMOPs, but they still suffer from the problems of easily getting trapped into local optimal solutions and low convergence. This paper introduces a multi-objective evolutionary algorithm based on decomposition and dynamic constraint-handling mechanism (MOEA/D-DCH) to tackle this issue. Firstly, the dynamic constraint-handling mechanism divides the search modes into the unconstrained search mode and the constrained search mode, which are dynamically adjusted by the generation number and the proportion of feasible solutions in the population. This mechanism could lead to a faster convergence than the traditional constraint-handling mechanisms. For the constrained search mode, an improved epsilon constraint-handling method is used to enhance the diversity of the population. Then, an individual update mechanism based on the best feasible solution of each sub-problem is designed to update the feasible individuals for maintaining the convergence of the feasible solutions. Finally, MOEA/D-DCH dynamically regulates the parameters of the differential evolution operator to enhance the local search ability. Experiments on 21 benchmark test functions are conducted to test MOEA/D-DCH and five other typical CMOEAs. Meanwhile, a real-world problem is employed to evaluate the practical performance of MOEA/D-DCH. MOEA/D-DCH achieves significantly better results than the other five algorithms on most of the test problems. The results indicate the effectiveness and competitiveness of MOEA/D-DCH for solving CMOPs.