基于强化学习的多雷达抗干扰算法研究  

A Study on Reinforcement Learning for Multi-radar Coexistence Anti-jamming

作　　者：智永锋[1] 邱璐莹 张龙[1] 高红岗 师浩博 ZHI Yongfeng;QIU Luying;ZHANG Long;GAO Honggang;SHI Haobo(Unmanned System Research institute,Northwestern Polytechnical University,Xi′an Shaanxi 710072,China;School of Civil Aviation,Northwestern Polytechnical University,Xi′an Shaanxi 710072,China)

机构地区：[1]西北工业大学无人系统技术研究院,陕西西安710072 [2]西北工业大学民航学院,陕西西安710072

出　　处：《现代雷达》2024年第2期131-137,共7页Modern Radar

摘　　要：针对多雷达系统在受到环境的扫频干扰下无法工作的问题,研究了基于深度强化学习的多雷达共存抗干扰算法。文中将环境划分为多个子频段,对干扰占用频段过程进行建模,用马尔可夫模型对多雷达系统进行建模;对双深度Q网络(Double DQN)强化学习算法进行改进,与门控单元循环神经网络相结合,使之能处理依赖于长时间序列的干扰问题;提出了基于门控循环记忆的深度确定性策略强化学习算法,针对Double DQN强化学习中的网络臃肿和行动集合较大的问题进行了改进,采用直接输出行动策略,有效降低了网络复杂度。实验仿真结果表明,在多雷达存在的情况,该算法通过避开存在干扰的频点,不仅能够有效降低来自外界的干扰,还能减少己方雷达相互之间的干扰。Aiming at the problem that the multi-radar system cannot work under the frequency sweep interference of the environment,a multi-radar co-existence anti-jamming algorithm based on deep reinforcement learning is studied.In this paper,the environment is divided into multiple sub-bands,the process of jamming occupying the frequency band is modeled,and the multi-radar system is modeled with Markov model.The double deep q-network(DQN)reinforcement learning algorithm is improved,and combined with the gating unit cyclic neural network,so that it can deal with the interference problem that depends on long time series.The deep deterministic strategy reinforcement learning algorithm based on gated recurrent memory is proposed,which improves the network overstaffing and large action set in double DQN reinforcement learning,and adopts the direct output action strategy to effectively reduce the network complexity.The simulation results show that in the case of multiple radar,the algorithm can not only reduce the interference from the outside world,but also reduce the interference between our own radars by avoiding the frequency points with interference.

关 键 词：多雷达系统 深度强化学习 抗干扰 马尔可夫模型 门控循环单元 

分 类 号：TN973[电子电信—信号与信息处理]