欢迎访问《兵工学报》官方网站,今天是

兵工学报 ›› 2022, Vol. 43 ›› Issue (3): 720-728.doi: 10.12382/bgxb.2021.0113

• 论文 • 上一篇    

装甲装备维修器材需求数据模拟

张磊1, 李世民2, 康淑瑰1, 王铁宁3, 郭猛超4   

  1. (1.山西大同大学 数学与统计学院, 山西 大同 037009; 2.63963部队, 北京 100072;3.陆军装甲兵学院, 北京 100072; 4.陆装驻西安地区军事代表局, 陕西 西安 710032)
  • 上线日期:2022-04-07
  • 作者简介:张磊(1983—), 男, 讲师,博士。E-mail: dtu_zhanglei@163.com
  • 基金资助:
    国家自然科学基金项目(11871314); 山西省高等学校科技创新项目(2020W111)

Data Simulation of Maintenance Material Demand for Armored Equipment

ZHANG Lei1, LI Shimin2, KANG Shugui1, WANG Tiening3, GUO Mengchao4   

  1. (1.College of Mathematics and Statistics Science, Shanxi Datong University, Datong 037009, Shanxi, China; 2.Unit 63963 of PLA, Beijing 100072, China; 3.Army Academy of Armored Force, Beijing 100072, China; 4.Army Equipment Department Military Representative Bureau in Xi'an, Xi'an 710032, Shaanxi, China)
  • Online:2022-04-07

摘要: 为解决小样本条件下运用大数据驱动模型预测装备维修器材需求效果不理想的问题,提出一种新的数据模拟扩充算法。通过将不同时间段的器材累积消耗值作为相似度量改进了AP聚 类算法,并对数据进行迭代聚类。结合改进后聚类算法的特点,将聚类为同一类别数据的各报告期器材相对消耗值考虑为正态分布数据,进而确定待预测数据各分量的正态分布数字特征。运用数学软件生成大量随机模拟样本构建训练集并进行预测。数值算例结果表明,通过本文方法生成的模拟样本可有效提升大数据驱动预测模型用于小样本器材需求预测的效果。随着模拟样本数据数量的增多,不同预测模型的预测结果稳定于同一预测值附近,有效提高大数据驱动模型预测结果的可信性。

关键词: 器材需求预测, 小样本, 数据模拟, 数据挖掘

Abstract: The predictive effect of the big-data-driven prediction model on the consumption demand of equipment maintenance materials with small sample size is not ideal. A data simulation expansion algorithm is proposed. The AP clustering algorithm is improved by taking the accumulated equipment consumption at different time stages as similarity measure, and the data at different time stages are clustered by iteration. Each relative equipment consumption in the report stage, in which the data are clustered as the same category, is considered as normally distributed data. The normal distribution characteristics of each component of a vector to be predicted are calculated, and a large number of random samples are generated by using mathematical software as the simulation training samples of the data-driven prediction model. The numerical results show that the simulation training samples generated by this method can effectively improve the predictive effect of the big-data-driven prediction model with few samples, and with the increase in the number of simulation sample data, the predicted results of different prediction models are stable near the same predicted value, which effectively improves the credibility of the predicted results of the data-driven model.

Key words: equipmentdemandprediction, smallsamplesize, datasimulation, datamining

中图分类号: