航空兵器2025，Vol.32Issue(5)：64-71,8.DOI:10.12132/ISSN.1673-5048.2025.0073
基于多智能体深度强化学习的高炮反无人机算法
Multi-Agent Deep Reinforcement Learning for Anti-Aircraft Artillery Systems in Counter-UAV Operations
摘要
Abstract
To address the issues of low engagement efficiency and insufficient adaptability in current coun-ter unmanned aerial vehicle(UAV)artillery systems,this paper proposes a situational-fused hierarchical multi-objective multi-agent reinforcement learning for counter-UAV systems.Firstly,within the context of counter-UAV artillery fire engagement systems,the problem of the counter-UAV defense scenario is formally defined.Secondly,the counter-UAV artillery task is formulated as a Markov decision process.Decision-making agents are constructed,and their state space,action space,and reward function are defined.Specifically,to enhance the agents' global situational awareness,multi-source situational information is fused into the state space.Addi-tionally,considering the characteristics of artillery counter-UAV engagement and fire control delay,a hierarchi-cal multi-objective reward mechanism is designed to guide the agents' decision-making process.Finally,the a-gents are trained using a deep multi-agent reinforcement learning approach based on monotonic value function factorization and validated within a simulation environment.Experimental results demonstrate that across three distinct counter-UAV scenarios,the proposed algorithm achieves task completion rates of 86％,88％,and 78％,respectively.This represents an average improvement of 48.9％compared to other prevalent MADRL al-gorithms.The results significantly enhance the engagement efficiency and battlefield adaptability of the counter-UAV artillery system,providing an effective AI-driven solution for countering UAV threats with artillery.关键词
高炮反无人机/态势信息/层次化多目标奖励/多智能体/深度强化学习/马尔可夫决策Key words
anti-aircraft artillery counter-UAV/situational information/hierarchical multi-objective re-ward/multi-agent/deep reinforcement learning/Markov decision分类
武器工业引用本文复制引用
胡家威,代昌华,祁万龙,陈志恒,王震,樊浡昊,郑欣磊,唐杰..基于多智能体深度强化学习的高炮反无人机算法[J].航空兵器,2025,32(5):64-71,8.基金项目
智能基础算法与技术应用团队计划(L2024-CXNL-KJRCTD-KJTD-0006) （L2024-CXNL-KJRCTD-KJTD-0006）
