欢迎访问《兵工学报》官方网站,今天是 分享到:

兵工学报 ›› 2023, Vol. 44 ›› Issue (1): 176-192.doi: 10.12382/bgxb.2022.0192

所属专题: 特种车辆理论与技术

• • 上一篇    下一篇

基于多层次LSTM网络的多智能体攻防效能动态预测模型

丁伟, 明振军*(), 王国新, 阎艳   

  1. 北京理工大学 机械与车辆学院, 北京 100081
  • 收稿日期:2022-03-24 上线日期:2022-07-22
  • 通讯作者:
  • 基金资助:
    国家自然科学基金项目(518005xxxx)

Dynamic Prediction Model Based on Multi-level LSTM Network for Multi-agent Attack and Defense Effectiveness

DING Wei, MING Zhenjun*(), WANG Guoxin, YAN Yan   

  1. School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China
  • Received:2022-03-24 Online:2022-07-22

摘要:

针对多智能体攻防体系存在多层次耦合、无规律涌现等特征而导致难以准确预测作战效能的问题,构建一种基于多层次长短时时间记忆(LSTM)网络的多智能体攻防效能动态预测模型。明确多智能体攻防的总体框架和作战流程,通过多主体NetLogo平台模拟红蓝智能体攻防对抗过程,以获取群体结构和作战效能在不同个体决策下的多层次演化数据。利用善于处理时序特征的LSTM网络来表征个体决策、群体结构和作战效能三层间的函数映射,并基于该映射关系进一步预测未来攻防作战效能与进程。上述建模方法已在多组仿真中证实了其可行性与有效性。实验结果表明,所建模型的作战效能预测误差仅在7%以内,对多智能体攻防的作战指挥和体系建设具有指导意义。

关键词: 多智能体攻防, 作战效能, NetLogo平台, 长短时记忆网络, 预测模型

Abstract:

Considering the difficulty of accurately predicting the operational effectiveness of multi-agent attack and defense (MAAD) systems due to multi-level coupling and irregular emergence, a dynamic prediction model based on multi-level LSTM network is constructed. The overall framework and operational process of the MAAD are clarified. Then, the multi-agent NetLogo platform is used to simulate the attack and defense confrontation process of red and blue agents in order to obtain multi-level evolutionary data of population structure and operational effectiveness when different decisions are made. On this basis, long short-term memory (LSTM) networks, which are effective in processing temporal features, are adopted to characterize the function mapping among the three layers of “individual decision, population structure, and operational effectiveness”, and to further predict the future attack and defense operational effectiveness and process based on the mapping relationship. The above modeling method has been proved to be feasible and effective in multiple simulations. The experimental results show that the prediction error of the model is only within 7%, which can serve as an effective guide for operational command and system development in MAAD systems.

Key words: multi-agent attack and defense, operational effectiveness, NetLogo platform, long short-term memory networks, prediction model

中图分类号: