An Adaptive Stochastic Ranking Mechanism in MOEA/D for Constrained Multi-objective Optimization
Weiqin Ying, Weipeng He, Yan-Xia Huang, Da-Tong Li, Yu Wu · 2016
Stochastic ranking is a promising technique for handling constraints in constrained evolutionary optimization. In stochastic ranking, the comparison between a pair of individuals is randomly based on either their objective values or degrees of constraint violation according to a constant probability parameter. In this paper, an adaptive stochastic ranking mechanism is presented to solve constrained multi-objective optimization problems (CMOPs) more effectively in the framework of the multi-objective evolutionary algorithm based on decomposition (MOEA/D). This mechanism dynamically controls the probability parameter according to both the current evolutionary stage and the difference between the degrees of constraint violation of individuals. Additionally, an elitist archiving scheme is introduced to preserve the so-far best feasible solution for each scalar sub-problem in the MOEA/D. Experimental results on CTP-series test instances show better and competitive performance of the proposed algorithm in terms of both convergence and diversity of solutions.