系统仿真学报 ›› 2024, Vol. 36 ›› Issue (3): 608-624.doi: 10.16182/j.issn1004731x.joss.22-1256

• 论文 • 上一篇    下一篇

面向兵棋推演复盘分析的机器学习数据集构建

张大永1(), 杨镜宇2, 马骏2, 宋晨烨1   

  1. 1.国防大学 研究生院,北京 100091
    2.国防大学 联合作战学院,北京 100091
  • 收稿日期:2022-10-20 修回日期:2022-12-09 出版日期:2024-03-15 发布日期:2024-03-14
  • 第一作者简介:张大永(1986-),男,博士生,研究方向为联合作战体系能力分析与评估。E-mail:dy311313@163.com

Construction of Machine Learning Data Set for Analyzing the Replay of the Wargaming

Zhang Dayong1(), Yang Jingyu2, Ma Jun2, Song Chenye1   

  1. 1.Graduate School, NDU of PLA, Beijing 100091, China
    2.College of Joint Operation, NDU of PLA, Beijing 100091, China
  • Received:2022-10-20 Revised:2022-12-09 Online:2024-03-15 Published:2024-03-14

摘要:

运用机器学习进行兵棋推演复盘分析,首先要解决的是数据集构建问题。由于机器学习对数据结构的规范化要求,以及算力和存储限制,通过兵棋推演数据构建机器学习数据集,在如何描述兵棋推演状态,如何描述推演过程,如何处理高维数据,如何数据保真等方面,还面临不少问题。针对此类问题,构建了兵棋推演过程数据向机器学习数据集映射模型,在总体框架上对数据集构建的映射流程、态势描述数据范围和数据统计计算规则进行规范,并从时间关联数据、地理空间关联数据和高维数据降维3个视角设计针对性处理方法,以保证构建数据集的数据结构统一、高维数据降维需求和数据集保真要求。通过数据集构建实验进行了验证,结果表明:在时间分辨率和地理空间分辨率适中情况下,所构建数据集映射模型,既能较好对兵棋推演高维数据进行降维,又能较好防止构建的数据集失真。

关键词: 兵棋推演, 复盘分析, 机器学习, 数据集, 构建方法

Abstract:

The first problem to be solved in the application of machine learning to the analysis of the replay of the wargaming is the construction of data sets. Due to the standardization requirements of machine learning for data structure, as well as the limitations of computing power and storage, building a machine learning data set through the wargaming data still faces many problems in terms of how to describe the wargaming situation, how to describe the wargaming process, how to handle high-dimensional data, and how to prevent data distortion. To solve these problems, this paper constructs a mapping model from the wargaming process data to the machine learning data set, standardizes the mapping process, situation description data range, and data statistics calculation rules of data set construction from the model framework, and designs targeted processing methods from three perspectives of time-related data, geospatial-related data, and high-dimensional data reduction, so as to ensure that the data structure of the data set is unified, and the dimension reduction requirements of high-dimensional data and the fidelity requirements of the data set are met. Through the data set construction experiment, it is verified that the data set mapping model constructed in this paper can not only reduce the dimension of high-dimensional data of the wargaming but also prevent the distortion of the constructed data set under the condition of moderate temporal resolution and geospatial resolution.

Key words: wargaming, replay analysis, machine learning, data set, construction method

中图分类号: