Large-Scale Multi-Objective Sensor-Weapon-Target Assignment Optimization: Benchmark Suites and Knowledge-Driven Heuristics
针对大规模多目标传感器-武器-目标分配问题，提出了包含复杂时空约束的基准套件SWTA_MOP1-5，以及利用拦截性矩阵知识驱动的稀疏初始化和不可行修复两种启发式策略，显著提升了八种先进算法的性能。
With the expanding deployment of unmanned swarms on the battlefield, the optimization of the Large-Scale Multi-Objective Sensor-Weapon-Target Assignment (LSMO-SWTA) problem has become increasingly critical to achieving effective defense operations. The core challenge lies in the combinatorial explosion inherent to this problem, resulting in a large-scale decision space where feasible regions satisfying stringent battlefield constraints are very sparse. However, existing studies often oversimplify the complex spatiotemporal constraints, revealing an urgent demand for specialized high-fidelity benchmark suites. To bridge this gap, we propose SWTA_MOP1-5, a benchmark suite capturing four configurable objectives (value, cost, timeliness, and risk) integrated with complex operational constraints, while supporting customizable problem scales. To address the challenges caused by large-scale decision spaces and complex constraints in LSMO-SWTA problems, we propose two knowledge-driven heuristics: sparse initialization and infeasible repair. By leveraging the target knowledge encoded in the interceptability matrices, the initialization strategy prunes the large-scale search space to quickly locate feasible regions, while the repair strategy intelligently rectifies infeasible offspring solutions prior to evaluation. To validate the effectiveness and generalization ability of the proposed strategies, we integrate them into eight state-of-the-art algorithms and conduct extensive experiments on 162 instances generated by SWTA MOP1-5. Experimental results demonstrate that the knowledge-driven variants outperform their original counterparts significantly, confirming that incorporating domain knowledge is pivotal for solving LSMO-SWTA problems.