摘要
针对多无人战车陆上突防作战时如何根据实时态势进行协同智能决策这一问题,结合多智能体无人战车突防作战过程建立马尔可夫(MDP)模型,并基于多智能体深度确定性策略梯度算法(Multi-agent Deep Deterministic Policy Gradient,MADDPG)提出多无人战车协同突防决策方法。针对多智能体决策时智能体策略变化互相影响的问题,通过在算法的AC结构中引入自注意力机制,使每个智能体进行决策和策略评估时更加关注那些对其影响较大的智能体;并采用自注意力机制计算每个智能体的回报权值,按照每个智能体自身贡献进行回报分配,提升了战车间的协同性;最后通过在想定环境中进行实验,验证了多战车协同突防决策方法的有效性。
Aiming at the problem of how to make intelligent cooperative decision according to the real-time situation in the land penetration operation of multi-vehicle,combined with the process of multi-agent unmanned vehicle penetration operation,Markov(MDP)model is established,and based on the multi-agent depth deterministic strategy gradient algorithm,the decision method of multi-unmanned vehicle collaborative penetration is proposed.In order to solve the problem of mutual influence of multi-agent decision-making agents'policy changes,an attention mechanism is introduced in AC structure of the algorithm to make each agent pay more attention to those agents that have greater influence on the decision-making and policy evaluation.And the self-attention mechanism is used to calculate the reward weight of each agent,the reward distribution is carried out according to the contribution of each agent,which improves the cooperation of the war shop.Finally,the effectiveness and superiority of the multi-vehicle collaborative penetration decision-making method are verified by experiments in a given environment.
作者
殷宇维
王凡
丁录顺
边金宁
YIN Yuwei;WANG Fan;DING Lushun;BIAN Jinning(Jiangsu Automation Research Institute,Lianyungang 222006)
出处
《指挥控制与仿真》
2025年第3期40-49,共10页
Command Control & Simulation
关键词
深度强化学习
多无人战车协同突防
多智能体深度确定性策略梯度
自注意力机制
deep reinforcement Learning
multiple unmanned vehicles coordinated penetration
multi-agent depth deterministic policy gradient
self-attention mechanism
作者简介
殷宇维(1997-),男,助理工程师,研究方向为人工智能;王凡(1970-),男,研究员。
