基于DEVS原子模型的智能体离散仿真构建方法  

Constructing the Agent Discrete Simulation Based on DEVS Atomic Model

作　　者：王霄汉 张霖[1,2] 赖李媛君[1,2] 谢堃钰 胡听春 Wang Xiaohan;Zhang Lin;Laili Yuanjun;Xie Kunyu;Hu Tingchun(Beihang University,Beijing 100191,China;Engineering Research Center of Complex Product Advanced Manufacturing Systems,Ministry of Education,Beijing 100191,China)

机构地区：[1]北京航空航天大学,北京100191 [2]复杂产品先进制造系统教育部工程研究中心,北京100191

出　　处：《系统仿真学报》2022年第2期191-200,共10页Journal of System Simulation

基　　金：国家重点研发计划(2018YFB1701600)。

摘　　要：智能体由于自身交互行为与学习行为的复杂性,难以直接被建模和仿真。针对智能体离散仿真中的常见问题,借助DEVS(discrete event system specification)原子模型的事件转移机制表示智能体的交互与学习过程,通过对智能体交互模式、多状态外部事件转移控制、端口连接模式、以及强化学习事件转移表示等原子模型下智能体建模技术的介绍,给出了基于DEVS原子模型的智能体离散仿真构建方法。在网格世界与倒立摆2个环境中进行仿真验证,实验结果证明了提出方法在构建智能体交互行为和学习行为的可行性和有效性。Agents are difficult to be directly modeled and simulated due to the complexity of their own interaction and learning behaviors. Aiming at the common problems in the discrete simulation of the agent, the event transfer mechanism of the discrete event system specification(DEVS) atomic model is applied to express the interaction and learning of an agent. Through the interaction mode of the agent,the transfer control of multi-state external events, the port connection mode, as well as the introduction of reinforcement learning event transfer representation, a discrete simulation construction method of the agent based on the DEVS atomic model is provided. The simulation verification is carried out in the grid world and the cart-pole environment. The experimental results prove the feasibility and effectiveness of the proposed method in constructing the interactive and learning behaviors of the agent.

关 键 词：智能体 DEVS 离散仿真 强化学习 状态转移 原子模型 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]