A constrained multi-objective evolutionary algorithm based on online problem identification and separate handling

Abstract

Solving constrained multi-objective optimization problems (CMOPs) is challenging because it requires optimizing multiple conflicting objectives and satisfying constraints simultaneously. In recent years, to better handle CMOPs, constrained multi-objective evolutionary algorithms (CMOEAs) based on the strategy of identifying problem types have been proposed. However, their performance remains limited due to low identification accuracy and inefficient constraint-handling techniques. In this work, a CMOEA based on online problem identification and separate handling, named CMOEA-IH, is proposed. First, to improve the accuracy of problem identification, an online problem identification strategy is proposed to identify the problem type during the entire evolution process. Second, based on the identified type, different constraint-handling techniques are employed by simultaneously considering the information from the unconstrained Pareto front and the Pareto front of single constraints. Finally, experimental results on 5 test suites and 3 real-world problems demonstrate that our proposed algorithm is more competitive in comparison with 10 state-of-the-art CMOEAs.