系统工程与电子技术 ›› 2018, Vol. 40 ›› Issue (1): 114-118.doi: 10.3969/j.issn.1001-506X.2018.01.17

• 系统工程 • 上一篇    下一篇

基于插值-拟合-迁移学习算法的机载设备故障概率预测

顾涛勇, 郭建胜, 李正欣, 王健, 王腾蛟   

  1. 空军工程大学装备管理与安全工程学院, 陕西 西安 710051
  • 出版日期:2018-01-08 发布日期:2018-01-08

Airborne equipment fault probability prediction based on interpolation-fitting-transfer learning algorithm

GU Taoyong, GUO Jiansheng, LI Zhengxin, WANG Jian, WANG Tengjiao   

  1. Equipment Management and Security Engineering College, Air Force Engineering University, Xi’an 710051, China
  • Online:2018-01-08 Published:2018-01-08

摘要:

针对不同工作环境的机载设备故障概率预测问题,提出自适应权重的插值-拟合-迁移学习 (interpolation-fitting-transfer learning, ITF)算法。算法根据数据量和数据特征(分布相似度和信息熵)对插值、拟合、迁移学习赋予一定的权重进行线性组合。插值和拟合方法可以对故障频率进行平滑,而迁移学习可以规避数据贫化所引起的预测风险。分析该方法的可行性,通过仿真实例展示算法在预测准确度上的优势,并讨论算法中仍待解决的问题和下一步的工作。

Abstract:

To solve the multi-operating environment airborne equipment fault probability prediction problem, a self-adaptive weighted interpolation-fitting-transfer learning (ITF) algorithm is proposed. On the basis of data size and data feature (distribution similarity and information entropy), the algorithm adjusts the weight of interpolation, fitting and transfer learning. The conventional interpolation and fitting method can smooth the fault frequency curve, and transfer learning can reduce the prediction risk caused by data dilution. The analysis and simulation demonstrate that the ITF algorithm is acceptable in time complexity and has higher prediction accuracy.