基于深度学习的高速铁路工务安全指数预测技术  

Deep-Learning-Based High-speed Railway Permanent-way Safety Index Prediction Technology

作　　者：柴雪松 凌烈鹏 周游 王萌瑶 CHAI Xuesong;LING Liepeng;ZHOU You;WANG Mengyao(Railway Engineering Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;School of Mathematics and Statistics,Beijing Jiaotong University,Beijing 100091,China)

机构地区：[1]中国铁道科学研究院集团有限公司铁道建筑研究所,北京100081 [2]北京交通大学数学与统计学院,北京100091

出　　处：《中国铁路》2022年第12期94-98,共5页China Railway

摘　　要：高速铁路工务安全指数(HRPSI)反映了高速铁路工务故障和事故的发生状况,对其进行规律验证与预测对于高速铁路工务专业进行安全评估和预测具有非常重要的现实意义。基于高速铁路10周年工务安全指数数据,构建2种深度学习的时间序列预测模型。利用皮尔森系数预测模型的有效性证明构建2种模型的有效性。其中门控循环单元(GRU)预测方法效果更好,训练集和测试集的皮尔森系数分别为0.9371和0.9221,可有效预测工务安全指数变化趋势。High‑speed Railway Permanent‑way Safety Index(HRPSI)reflects the occurrence of HSR permanent way faults and accidents,and its regular verification and prediction are of great practical significance for the safety evaluation and prediction for HSR permanent way.Based on the HRPSI data of the last ten years,two time-series prediction models for deep learning are established.The validity of the Pearson coefficient prediction model is used to prove the validity of the two models.Among them,the prediction method GRU has better effect,and the Pearson coefficients of training set and test set are 0.9371 and 0.9221 respectively,which proves that it can effectively predict the change trend of HRPSI.

关 键 词：高速铁路 深度学习 统计分析 高速铁路工务安全指数 门控循环单元 长短期记忆神经网络 

分 类 号：U216[交通运输工程—道路与铁道工程]