基于LSTM的动车组故障率预测模型

LSTM-based Model for Predicting Failure Rate of EMUs

动车组故障率趋势波动较大,其机理因素较为复杂:既有源头质量、养护维修问题产生的起伏,又有线路条件、气候等环境因素导致的故障率波动,同时涉及动车组生命周期内大量转配问题导致其运行环境的改变。由于很难量化这些因素,采用传统数理拟合方式描述动车组安全规律特征难度较大。提出1种长短记忆(LSTM)神经网络模型,用数据驱动的方式对高速动车组安全规律进行建模,以期预测未来周期的故障率数据。该模型能通过合理泛化训练,一定程度上通过对各系统安全规律数据的收集,掌握整车故障率的发展趋势,可为动车组运维提供数据支撑。

The failure rate of EMUs changes greatly,with complex causing mechanisms.It is caused not only by source quality and maintenance problem,but also by environmental factors,such as track condition and climate.Meanwhile,frequent transfer in the life cycle of EMUs makes the operation environment change.These causes are difficult to be quantified.Therefore,it is difficult to define the characteristics of safety rules by using the conventional method of mathematical fitting.A long-short-term memory(LSTM)neutral network model is proposed to model the safety rules of high speed EMUs in a data-driven way for the purpose of predicting the failure rate data in future cycles.The development tendency of the failure rate of the complete vehicle can be described with this model,receiving proper generalization exercise and collecting system safety rules data,which can provide data support for EMU operation and maintenance.