Constrained Multi-Objective Optimization With Constraint Priority

Constrained multi-objective problems (CMOPs) are tricky, because it is difficult to handle multiple objectives and constraints simultaneously. Most existing algorithms perform well on CMOPs with a single constraint or multiple simple constraints, but their performance on CMOPs with multiple complex constraints often deteriorates. Generally, for CMOPs, it is much easier to deal with a single constraint than with multiple constraints. Therefore, this paper proposes a strategy for identifying the constraint priority, and then handles the constraints one by one according to the priority. Specifically, for each constraint, an associated population is first established to exclusively consider this constraint, and then the corresponding infeasibility rates of all the constraints are calculated to capture the constraint priority. After that, based on the constraint priority, the roulette selection method is used to determine which constraint should be handled first, which ensures the constraint with high priority has more chances to be handled. Experimental results on five benchmark suites and three real-world problems show that our approach has significant advantages over five state-of-the-art algorithms.