面向智能运维的轨道交通转辙机模拟数据生成器设计与验证
邹劲柏1魏诗燕1刘江2沙泉1吴杰3季国一1
Design and Verification of Simulation Data Generator for Rail Transit Switch Machine Oriented to Intelligent Operation and Maintenance
ZOU Jinbai1WEI Shiyan1LIU Jiang2SHA Quan1WU Jie3JI Guoyi1
-
作者信息:1.上海应用技术大学轨道交通学院，201400，上海
2.北京交通大学电子信息工程学院，100044，北京
3.上海地铁维护保障有限公司通号分公司，200235，上海
-
Affiliation:1.School of Railway Transportation, Shanghai Institute of Technology, 201400, Shanghai, China
2.School of Electronic and Information Engineering, Beijing Jiaotong University, 100044， Beijing, China
3.Telecom ＆ Signaling Branch, Shanghai Metro Maintenance Support Co., Ltd., 200235, Shanghai, China
-
关键词:
-
Key words:
-
DOI:10.16037/j.1007-869x.2025.01.034
-
中图分类号/CLCN:U213.6+1
-
栏目/Col:研究报告
摘要:
［目的］由于轨道交通各类设备的故障数据难以轻易获取，导致在开展故障诊断与预测等机器智能算法研究时缺乏充足的数据支持。为了满足轨道交通智能运维对大量训练数据的迫切需求，有必要设计轨道交通转辙机模拟数据生成器并对其进行验证。［方法］对S700K型转辙机正常动作与缓变性故障的功率曲线特征进行了分析，并探讨了故障发生原因。通过对比两种模拟数据生成方法，基于Borderline-Smote算法设计出转辙机模拟数据生成器，搭建转辙机模拟数据生成器平台，利用LSTM（长短期记忆）预测模型学习功率数据的时间序列特征,对生成的缓变性故障功率数据的峰值因子、标准差和方差等3个特征进行试验。［结果及结论］通过转辙机模拟数据生成器生成的功率数据训练出的LSTM预测模型，可以预测出S700K型转辙机的功率变化趋势。通过对比LSTM预测模型与周期性复制法计算得到的峰值因子、标准差、方差的均方根误差分别为0.3355、0.0239和0.0241，误差较小，证明转辙机模拟数据生成器的真实性及可行性。
Abstracts:
［Objective］ Difficulty in easily obtaining fault data of various rail transit equipment leads to insufficient data to support the research on machine intelligence algorithms such as fault diagnosis and prediction. In order to meet the urgent need of intelligent rail transit operation and maintenance for a large amount of training data, it is necessary to design and verify the simulation data generator (hereinafter abbreviated as SD generator) of rail transit switch machine. ［Method］ The characteristics of S700K type switch machine power curves under normal operation and gradual fault conditions are analyzed, and causes of the faults are discussed. By comparing two simulation data generation methods, a switch machine SD generator is designed based on the Borderline-Smote algorithm. Through building a platform for SD generator, and using the time series features of learning the power data by LSTM (long short-term memory) prediction model, three characteristics of the generated gradual fault power data such as the crest factor, standard deviation, and variance are tested. ［Result & Conclusion］ The LSTM prediction model trained on the power data generated by SD generator can predict the power change trend of the S700K type switch machine. The root mean square errors of crest factor, standard deviation, and variance calculated by the LSTM prediction model and the periodic replication method are 0.335 5, 0.023 9, and 0.024 1 respectively. The relatively small errors prove the authenticity and feasibility of SD generator.
- 上一篇: 高铁物流运输模式及其可行性
- 下一篇: 超导电动磁浮技术完善交通系统产业链
过刊浏览
资料下载
友情链接