摘要
针对传统基于规则的建模方法存在规则建模有限性及行为描述不全面、难以应对作战过程中的未知事件等问题,建立了“感知行为-决策行为-物理行为”模型整体概念及基于规则与学习一体化的建模框架,实现了一种基于行为树-深度神经网络-深度强化学习(BDD)的作战实体认知行为一体化建模方法,包括基于深度神经网络的未知事件感知与特征识别、基于PER-DQN算法的未知事件处置行为决策,它们与行为树方法共同完成完整战场态势的处理。以战场侦察活动为例进行了建模验证与仿真分析,实验表明,上述方法拓展了传统方法作战实体认知行为的建模能力,可解决其行为描述有限性问题,可应用于复杂战场环境下作战实体行为建模。
Traditional rule-based modeling methods have some problems,such as limited rule modeling,incomplete behavior description,and difficulty in dealing with unknown events in combat.For these problems,an overall concept of"perception behavior,decision behavior,physical behavior"and a modeling framework based on the integration of rules and learning are established,and an integrated modeling method of cognitive behavior of combat entities based on behavior tree,deep neural network and deep reinforcement learning(BDD)is implemented,including unknown event perception and feature recognition based on deep neural network,and behavior decision of unknown event disposal based on PER-DQN algorithm,which together with behavior tree method complete the processing of complete battlefield situation.The behavior model is built by taking battlefield reconnaissance as an example.Experimental results show that this method expands the decision-making ability of the behavior tree,and can solve the problem of the limitation of behavior tree rule modeling.It can be applied to modeling the behavior model of combat entities in complex battlefield environments.
作者
曹朋飞
邸彦强
孟宪国
李兴德
CAO Peng-fei;DI Yan-qiang;MENG Xian-guo;LI Xing-de(Army Engineering University Shijiazhuang Campus,Hebei,Shijiazhuang 050003,China;Army Engineering University Communication NCO Academy,Chongqing 400035,China)
出处
《计算机仿真》
2024年第7期6-13,49,共9页
Computer Simulation