基于双层强化学习的干扰策略与干扰波形优化设计  

Optimization Design of Interference Strategy and Interference Waveform Based on Two-layer Reinforcement Learning

作　　者：辛祺 辛增献 马亮 辛升 陈涛[1] XIN Qi;XIN Zengxian;MA Liang;XIN Sheng;CHEN Tao(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,Heilongjiang,China;Shanghai Radio Equipment Research Institute,Shanghai 201109,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [2]上海无线电设备研究所,上海201109

出　　处：《制导与引信》2023年第4期35-41,共7页Guidance & Fuze

基　　金：国家自然科学基金(62071137);国防科技基础加强计划(2019-JCJQ-ZD-067-00);上海航天科技创新基金(SAST2022-063)。

摘　　要：针对干扰策略与干扰波形联合优化设计问题,提出了一种基于双层强化学习的干扰策略与间歇采样转发干扰波形人工智能优化设计方法。该方法通过建立基于双层强化学习的干扰决策模型,外层利用Q学习(Q-learning)算法,基于雷达工作模式识别对干扰策略进行人工智能优化,内层利用深度Q学习网络(deep Q-leaning network,DQN)对非均匀间歇采样转发干扰波形进行人工智能优化,从而将一个干扰策略与相干干扰波形优化的二维决策问题转换为两个一维决策问题。仿真实验表明:该模型对于未知且复杂的电磁环境具有良好的自适应能力,为多层强化学习网络应用于复杂干扰决策场景提供了一种可行的解决方案。Aiming at the problem of joint optimization design of interference strategy and interference waveform,an artificial intelligence optimization design method of interference strategy and intermittent sampling and forwarding interference waveform based on two-layer reinforcement learning was proposed.In this method,an interference decision-making model based on two-layer reinforcement learning was established,the outer layer employed the Qlearning algorithm,conducting artificial intelligence optimization on the basis of radar working mode recognition for interference strategies,meanwhile,the inner layer utilized a deep Q-learning network(DQN)for artificial intelligence optimization of non-uniform intermittent sampling and forwarding interference waveforms.Therefore,a two-dimensional decision problem of interference strategy and coherent interference waveform optimization design was transformed into two one-dimensional decision problems.Simulation experiments show that the model has good adaptive ability for unknown and complex electromagnetic environments,which provides a feasible solution for multi-layer reinforcement learning network to be applied to complex interference decision-making scenarios.

关 键 词：干扰策略 干扰波形 强化学习 深度Q学习网络 间歇采样转发干扰 

分 类 号：TN974[电子电信—信号与信息处理]