摘要
针对基于深度强化学习方法解决实际博弈对抗序列规划问题中存在的探索-利用矛盾、奖赏信号稀疏、数据利用率低、难以稳定收敛等问题,分析了基于知识的学习型智能生成模式,提出基于知识驱动的方法,从用规则教、从数据中学、用问题引导等方面构建了智能博弈对抗行动序列规划模型,为提升探索-利用效率、精准奖励函数、加速算法收敛提供了理论支撑。对基于强化学习的智能博弈对抗问题求解的难点问题进行了讨论,指出下一步深度强化学习算法走向实用的发展方向。
Aiming at the problems of conflict between exploration and utilization,sparse reward signals,low data utilization rate,and difficulty in stable convergence in solving the practical course of action planning for Intelligent Game Confrontation based on deep reinforcement learning.The knowledge-based type-learning intelligent generation mode is analyzed,and the knowledge driven method is proposed.The course of planning model of intelligent game confrontation from the aspects of rule-based teaching,data-based learning and problem-based guidance and other aspects is constructed,which provides theoretical support for improving the exploration utilization efficiency,accurate reward function and accelerating algorithm convergence.The difficult problems of solving the intelligent game confrontation problem based on reinforcement learning are discussed,and the more practical development direction of the next step deep enforcement learning algorithm is pointed out.
作者
陈希亮
曹雷
康凯
李晨溪
CHEN Xiliang;CAO Lei;KANG Kai;LI Chenxi(College of Command and Control Engineering,Army Engineering University,Nanjing 210007,China;Unit 31108 of PLA,Nanjing 210007,China)
出处
《指挥与控制学报》
CSCD
北大核心
2024年第4期509-515,共7页
Journal of Command and Control
基金
国家自然科学基金(62273356)资助。
关键词
深度强化学习
博弈对抗
知识驱动
行动序列规划
deep reinforcement learning
intelligent game confrontation
knowledge driven
course of action planning
