基于脉冲神经网络的指挥智能体技术  

Command Agent Technology Based on Spiking Neural Network

作　　者：王栋 赵彦东 陈希飞 WANG Dong;ZHAO Yandong;CHEN Xifei(North Automatic Control Technology Institute,Taiyuan 030006,China)

机构地区：[1]北方自动控制技术研究所,太原030006

出　　处：《火力与指挥控制》2024年第5期36-43,51,共9页Fire Control & Command Control

摘　　要：针对现有智能体技术应用于军事指挥控制领域中时存在计算资源需求高、奖励值稀疏、收敛速度慢、推理效果差的问题,提出了一种基于脉冲神经网络(spiking neural network,SNN)和分层强化学习的指挥智能体技术。基于分层强化学习思想对军事指挥智能体进行建模,利用SNN构建智能体决策模型;通过ANN-SNN转换的学习算法获得基于SNN的指挥智能体;基于“墨子”兵棋推演软件开展对比试验,与现有智能体技术相比,提出方法对计算资源的需求较低,且具有较高的博弈对抗胜率。As for such problems existing in the application of current agent technology in the field of military command and control as high demand for computational resources,sparse reward values,slow convergence speed,and poor reasoning effects,a command agent technology based on Spiking Neural Network(SNN)and hierarchical reinforcement learning is proposed.Firstly,the military command agent is modeled based on the idea of hierarchical reinforcement learning,and the decision-making model of the agent is constructed with SNN.Then,the learning algorithm of ANN-SNN conversion is used to obtain the command agent based on SNN.Finally,a comparative experiment is carried out based on the“Mozi”wargaming software.Compared with the existing agent technology,the proposed method has lower demand for computational resources and higher win rate of game confrontation.

关 键 词：脉冲神经网络 分层强化学习 指挥智能体 ANN-SNN 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TP183[自动化与计算机技术—控制科学与工程]