An evolutionary multitasking algorithm based on k-nearest neighbors pre-selection strategy for constrained multi-objective optimization

Evolutionary algorithms for solving constrained multi-objective optimization problems have attracted considerable attention in recent years. These algorithms typically involve population initialization and evaluation, offspring generation, and environmental selection. However, many existing algorithms fail to improve solving efficiency due to the neglect of promising infeasible solutions. To address this issue, we propose a constrained multi-objective evolutionary algorithm that integrates a k-nearest neighbors (KNN)-based pre-selection strategy into the evolutionary multitasking framework (CMOEAKNN). Specifically, a KNN classifier is designed and trained to pre-select offspring with superior performance before performing environmental selection, thereby minimizing unnecessary evaluation efforts and retaining promising infeasible solutions, which improves the solving efficiency of the algorithm. The algorithm incorporates a reverse learning mutation strategy to improve population diversity and global exploration capability. The experiment results on three test suites and seven engineering application problems demonstrate that the proposed CMOEAKNN has significant competitiveness and superior performance compared to the other nine comparative algorithms.