Learning-Based Policy Optimization for Adversarial Missile-Target Assignment
提出一种基于深度强化学习的数据驱动策略优化方法，用于解决现代战争中导弹目标分配问题，能自适应生成高质量分配方案并预测资源需求。
The missile-target assignment (MTA) is a typical weapon-target assignment problem in Command and Control of modern warfare. Despite the significance of the problem, traditional algorithms still lack efficiency, solution quality, and practicability in the adversarial environment. In this article, we propose a data-driven policy optimization with deep reinforcement learning (PODRL) for the adversarial MTA. We design a comprehensive reward function to motivate the optimization of assignment policy. As such, the learned policy can implicitly model the penetration of missiles under an adversarial environment in a data-driven way. We also present a fair sample strategy to improve the sample efficiency and accelerate the policy optimization. Experimental results show that PODRL can adaptively generate satisfactory solutions in both small-scale and large-scale instances. Furthermore, we evaluate the effectiveness of PODRL in a multiobjective scenario. The result demonstrates that a well-optimized policy can achieve high-quality allocation and demand forecast of the missile resources simultaneously.