基于LSTM网络的轨道车辆基准轴速度预测方法  

Prediction Method of Reference Axle Speed of Rail Vehicles Based on LSTM Network

作　　者：孙卫兵 杨磊 方松[1] SUN Weibing;YANG Lei;FANG Song(School of Locomotive and Car,Nanjing Vocational Institute of Railway Technology,Nanjing Jiangsu 210031,China;CRRC Nanjing Puzhen Haitai Brake Equipment Co.,Ltd.,Nanjing Jiangsu 210031,China)

机构地区：[1]南京铁道职业技术学院机车车辆学院,江苏南京210031 [2]南京中车浦镇海泰制动设备有限公司,江苏南京210031

出　　处：《中国铁路》2025年第1期92-99,共8页China Railway

基　　金：江苏省轨道交通控制工程技术研究开发中心开放基金项目(KFJ2006);轨道交通装备数字化研究中心项目(KYPT2023001)。

摘　　要：滑行检测是列车制动系统防滑控制的关键技术,以真实轨道车辆制动系统的运行数据为样本进行特征分析,提出基于长短期记忆网络(LSTM)的列车基准轴速度预测方法。该方法根据车辆4个轴的实时速度及其邻近时刻的速度,对下一时间段的基准轴速度进行迭代预测。与常规基准轴速度估算方法相比,LSTM算法预测的基准轴速度在全轴滑行工况下更接近列车真实速度,可更早地检测到全轴滑行,有利于制动系统及时采取防滑控制措施或其他黏着控制,提高黏着利用率。Slide detection is a crucial technology for anti-slide control in train brake systems.Taking operational data from actual rail vehicle brake systems as samples for feature analysis,this study proposes a method for predicting the reference axle speed of trains based on Long Short-Term Memory(LSTM).The method iteratively predicts the reference axle speed for the next time period according to the real-time speeds of the vehicle's four axes and their speeds at adjacent time intervals.Compared to conventional methods for estimating reference axle speed,the prediction method based on the LSTM algorithm is closer to the actual train speed when all axles are sliding.This allows for earlier detection of this all-axle slide condition,enabling the brake system to implement anti-slide control measures or other adhesion controls in a timely manner,thereby improving adhesion utilization.

关 键 词：轨道车辆 基准轴速度 列车制动 长短期记忆网络 神经网络 滑行检测 黏着控制 

分 类 号：U271[机械工程—车辆工程]