基于进化强化联合算法的多域智能行动规划方法OA
Multi-domain intelligent action planning method based on joint Evolutionary and reinforcement learning algorithm
面向联合多域行动的智能规划,在多智能体深度强化学习方法的基础上提出一种进化策略-强化学习联合算法,利用多头注意力机制改进评价网络输入,结合进化策略改进信度分配机制,采用阶段学习方法改进权重训练策略.解决了多智能体系统在大规模决策空间下处理全局信息的能力下降以及强化学习智能体规模与训练效率受限的问题.仿真结果表明,该算法具有更好的学习效率、稳定性和适应性.
Aimed at intelligent planning for joint multi-domain operations,this study proposes an evolutionary strategy-rein-forcement learning joint algorithm based on multi-agent deep reinforcement learning.The method employs a multi-head atten-tion mechanism to enhance the input of the critic network,integrates evolutionary strategies to improve the credit assignment mechanism,and adopts a phased learning approach to optimize the weight training strategy.It addresses the challenges of degraded global information processing capability in multi-agent systems under large-scale decision spaces and the limitations of reinforcement learning in agent scalability and training efficiency.Simulation results demonstrate that the proposed algo-rithm achieves superior learning efficiency,stability and adaptability.
陈晨;刘海颖;侯宏楷;秦乐
南京航空航天大学,江苏 南京 210016南京应用数学中心,江苏 南京 211135南京航空航天大学,江苏 南京 210016南京航空航天大学,江苏 南京 210016
信息技术与安全科学
联合多域行动多智能体系统大规模决策空间进化算法
multi-domain operationsmulti-agent systemlarge-scale decision spaceevolutionary algorithm
《指挥控制与仿真》 2026 (5)
29-33,5
ZBYY研究基金
评论