基于分层多智能体强化学习的雷达协同抗干扰策略优化

打开文本图片集

关键词：雷达抗干扰策略；分层强化学习；多智能体系统；深度确定性策略梯度；稀疏奖励 中图分类号：TN974 文献标志码：A DOI：10.12305/j.issn.1001-506X.2025.04.07

Abstract： The sparsity of rewards in the decision-making process of radar collaborative anti-jamming makes it dificult for reinforcement learning algorithms to converge and for collaborative training.To address this issue，a hierarchical multi-agent deep deterministic policy gradient（H-MADDPG）algorithm is proposed. By accumulating sparse rewards，the convergence performance of the training process is improved，and the Harvard structure idea is introduced to separately store the training experiences of multi-agent to eliminate theconfusion in experience replay. In the simulations of two and four radars network simulation，under certain strong jamming conditions，the radar detection success rate is respectively increased by 15% and 30% compared to the multi-agent deep deterministic policy gradient（MADDPG）algorithm.

Keywords：radar anti-jamming；hierarchical reinforcement learning；multi-agent system；deep deterministic policy gradient （DDPG）；sparse reward

0 引言

组网雷达具有覆盖空间大、信息互通、资源共享的优点，与单雷达相比具有更大的覆盖面积和更好的探测性能。（剩余12403字）