基于强化学习的集群弹药引信协同起爆控制模型  

A cooperative detonation control model for cluster ammunition fuzes based on reinforcement learning

作　　者：张俊 付胜华 常博 李楚宝[2] ZHANG Jun;FU Shenghua;CHANG Bo;LI Chubao(Unit No.92942,Beijing 100161,China;Xi′an Institute of Electromechanical Information Technology,Xi′an 710065,China)

机构地区：[1]92942部队,北京100161 [2]西安机电信息技术研究所,西安710065

出　　处：《海军工程大学学报》2023年第4期88-92,112,共6页Journal of Naval University of Engineering

摘　　要：为实现集群弹药打击目标的高效毁伤并缩短引信起爆响应时间,提出了一种多智能体强化学习的集群弹药引信协同起爆控制模型。首先,利用弹药与目标交会距离、加速度、目标的运动参数,设置集群弹药奖励参数,实现与目标交会的最优路径与引信起爆时间;然后,为进一步提高模型的起爆控制精度,加入了近端策略优化方案,提高了短时间内最少能量消耗下多点协同起爆的准确率。仿真结果表明:该模型在设置较大的初始位置干扰后仍能够快速收敛,模拟集群弹药在接近目标的过程中始终保持较小的加速度,以此来使消耗能量最低,确保集群弹药引信对目标的可靠起爆。In order to achieve high efficiency damage to the target of cluster munitions and shorten the response time of fuze detonation,a multi-agent reinforcement learning model was proposed for coo-perative detonation control of cluster ammunition fuzes.Using the intersection distance,acceleration and motion parameters of ammunition and target,reinforcement learning reward parameters were set to achieve the optimal path and the detonation time of fuzes.Meanwhile,In order to improve the detonation control precision,a near-end strategy optimization scheme was added to improve the accuracy of multi-point cooperative detonation under the minimum energy consumption in a short time.The simulation results show that the model can still converge quickly after setting a large initial position interference,and that the simulated cluster munitions always maintain a small acceleration during the process of approaching the target,so as to minimize the energy consumption and ensure the reliable detonation of the cluster ammunition fuzes to the target.

关 键 词：集群弹药 引信 强化学习 协同起爆 

分 类 号：TJ43[兵器科学与技术—火炮、自动武器与弹药工程]