Abstract:

Reasonable selection of combat resources to form a sensor-weapon-target kill-chains plays an important role in air defense network operations. This paper studies sensor-weapon-target assignment （S-W-TA） under multiple constraints and multiple optimization indexes， and proposes an deep reinforcement learning based on allocation method. The mathematical model of S-W-TA problem is established， and the concept of kill chain advantage is used to integrate the traditional efficiency index. The deep Q network （DQN） training agent is used to solve the S-W-TA problem by deep reinforcement learning method for the first time. The simulation results show that the solution obtained by the deep reinforcement learning algorithm is superior to the rule-based allocation method widely used in engineering， and the reinforcement learning algorithm is more suitable for solving the S-W-TA problem with multiple constraints and multiple optimization indexes， and has certain engineering application value.

Key words: networked operation, sensor-weapon-target assignment（S-W-TA）, kill-chain, reinforcement learning, deep Q network

摘要：

合理选择作战资源组成“传感器-武器-目标”杀伤链在防空网络化作战中具有重要的意义，研究了多约束限制、多优化指标下的传感器-武器-目标分配（sensor-weapon-taget assignment，S-W-TA）问题，建立其数学模型，并提出一种基于深度强化学习的分配方法。分析S-W-TA问题对应的寻优指标，使用杀伤链有利度指标整合传统的效能指标；采用深度Q网络（deep Q network,DQN）方法训练智能体，使用深度强化学习类方法对S-W-TA问题进行求解。仿真结果表明：在杀伤链择优组网的过程中，深度强化学习算法所求得的解优于工程上广泛应用的基于规则的分配方法，强化学习类算法更适合解决多约束限制、多优化指标的S-W-TA问题，具有一定的工程应用价值。

关键词: 网络化作战, 传感器-武器-目标分配, 杀伤链, 强化学习, 深度Q网络