基于深度强化学习的干扰资源分配方法  

Jamming Resource Allocation Method Based on Deep Reinforcement Learning

作　　者：李健涛 王轲昕 刘凯 张天贤 LI Jiantao;WANG Kexin;LIU Kai;ZHANG Tianxian(School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China)

机构地区：[1]电子科技大学信息与通信工程学院,四川成都611731

出　　处：《现代雷达》2023年第10期44-51,共8页Modern Radar

基　　金：国家自然科学基金资助面上项目(61971109);国防科技创新特区支持项目(重点项目);中央高校其本科研业务费资助项目(ZYGX2018J009)。

摘　　要：针对干扰机群掩护目标突防组网雷达场景下的干扰资源分配的问题,提出了一种基于深度强化学习的干扰资源分配方法。该文将干扰资源分配模型描述为一个马尔可夫决策过程,并提出了一种基于动作密钥编码的双延迟深度确定性策略梯度(AKE-TD3)网络训练算法,将混合整数优化问题转化为连续变量优化问题,解决了算法难以收敛的问题。仿真结果表明,文中所设计的干扰资源分配方法对组网雷达有更好的干扰效果,且稳定性更高,有效地提升了干扰机群的作战性能。A deep reinforcement learning-based jamming resource allocation method is proposed to address the problem of resource allocation of multi-jammer for jamming netted radar system to cover target penetration.The jamming resource allocation model is described as a Markov decision process,and an action key encoding based double-delayed deep deterministic policy gradient(AKE-TD3)network training algorithm is also proposed,which transforms the mixed-integer optimization problem into a continuous-variable optimization problem to improve the convergence of the algorithm.Simulation results show that the jamming resource allocation method proposed in the paper has better jamming effectiveness and stability for netted radar systems.This method significantly enhances the combat performance of multi-jammer.

关 键 词：组网雷达系统 干扰资源分配 马尔科夫决策过程 深度强化学习 动作密钥编码 

分 类 号：TN972[电子电信—信号与信息处理]