Collaborative Orthogonal Learning for Constrained Multi-Objective Optimization
提出一种双种群竞争优化器，通过协同正交学习策略平衡可行性、收敛性和多样性，在复杂约束多目标问题上优于九种现有方法。
Balancing feasibility, convergence, and diversity is a fundamental challenge in constrained multi-objective optimization, particularly in landscapes with irregular or disconnected boundaries. Existing methods primarily exchange solution positions but neglect valuable convergence direction information contained within the evolutionary process, which may lead to inefficient and redundant searches for the constrained Pareto front (CPF). To address this, we propose a dual-swarm competitive optimizer with collaborative orthogonal learning (COL) strategy, which effectively decouples global exploration and diversity exploitation. Specifically, the main swarm performs trend learning to identify convergence directions from boundary and winner-loser interaction information, enabling global exploration through infeasible regions toward the CPF. Guided by the learned trends, the auxiliary swarm executes orthogonal learning to search complementary subspaces, which broadens the solution distribution while avoiding redundant searches, ensuring diversity exploitation capability. Additionally, a niche-guided subset selection strategy is introduced to maintain uniform distribution within the objective space through a three-level subset division mechanism based on local niche capacity. Extensive experiments on standard and extended benchmark instances demonstrate the robustness and superiority of our approach over nine state-of-the-art methods.