基于深度学习的机车故障预测方法研究  

Research on locomotive fault prediction method based on deep learning

作　　者：张立琛 李砾工 石勇 王忠福 ZHANG Lichen;LI Ligong;SHI Yong;WANG Zhongfu(CRRC Dalian Electric Traction R&D Center Co.,Ltd.,Dalian 116052,China)

机构地区：[1]中车大连电力牵引研发中心有限公司,辽宁大连116052

出　　处：《铁道机车与动车》2021年第S02期11-13,I0001,共4页Railway Locomotive and Motor Car

摘　　要：故障预测是提升机车安全保障能力、降低运维成本的重要手段。因目前解决预测问题主要采用人工特征和传统的预测算法,没有利用机车运行过程中产生的大量有价值的数据和专家知识,从而影响了预测的效果。提出一种基于神经网络的深度学习方法:首先分析设备状态数据与故障数据的关系;然后自动提取设备状态的历史数据和专家知识的特征,设计适合的网络结构,构建能够表达模型预测的故障和训练样本间差异的目标函数;最终以端到端的方式给出完整的预测解决方案。Fault prediction is an important means to improve the locomotive safety guarantee ability and reduce the operation and maintenance cost.At present,artificial features and traditional prediction algorithms are mainly used to solve the prediction problems,but a large number of valuable data and expert knowledge generated during locomotive operation are not used,which affects the prediction effect.A deep learning method based on neural network is proposed.Firstly,the relationship between the equipment state data and fault data is analyzed.Then,the historical data of the equipment state and the characteristics of expert knowledge are extracted automatically,the appropriate network structure is designed,and the objective function that can express the faults predicted by the model and the differences between the training samples is constructed.Finally,a complete prediction solution is presented in an end-to-end manner.

关 键 词：深度学习 故障预测 神经网络 目标函数 

分 类 号：U2[交通运输工程—道路与铁道工程] TP273[自动化与计算机技术—检测技术与自动化装置]