摘要
为研究有限作战指挥样本下的智能决策方法,针对作战决策经验难以表达和智能决策学习训练样本稀缺等问题,基于联合战役仿真推演环境,提出了一种基于生成对抗模仿学习的作战决策方法。该方法整合了作战决策经验表示与学习过程,在上层决策和底层动作分层的基础上,采用规则定义特定任务执行逻辑,并利用生成对抗模仿学习算法提升智能体场景泛化能力。在构设的典型对抗场景中,该方法达到了预期效果,算法训练收敛,智能体输出决策合理。实验结果初步表明,生成对抗模仿学习作为一种智能作战决策方法,具有进一步研究价值。
To study the intelligent decision making methods under limited decision samples,aiming at the problems that operational decisionmaking experience is difficult to express and the training samples for intelligent decision learning are limited,based on the joint operational simulation and drill environment,a decisionmaking method based on generative adversarial imitation learning is proposed.This method integrates the operational decisionmaking experience representation and learning process.On the basis of highlevel decisionmaking and lowlevel action,rule definitions are used to specify the logic of task execution,and generative adversarial imitation learning algorithms are utilized to improve the generalization ability of intelligent agents in scenarios.This method achieved expected results in the constructed typical adversarial scenarios.The algorithm training converged and the decisions output by the intelligent agent are reasonable.Preliminary experimental results indicate that generative adversarial imitation learning,as an intelligent operational decisionmaking method,has value for further research.
作者
李东
许霄
吴琳
LI Dong;XU Xiao;WU Lin(College of Joint Operation,National Defense University,Beijing 100091,China)
出处
《指挥控制与仿真》
2024年第2期18-23,共6页
Command Control & Simulation
基金
国家自然科学基金(62006235)。
关键词
智能决策
作战决策
基于规则的方法
生成对抗模仿学习
intelligent decision-making
operational decision-making
rule-based method
generative adversarial imitation learning
作者简介
李东(1987-),男,工程师,研究方向为军事智能决策;许霄(1987-),男,工程师。
