列车运行牵引制动故障分类的随机森林算法设计与验证
Design and Verification of Random Forest Algorithm for Traction/Braking Fault Classification in Train Operation
雷文丞 1郭静 2高士根3
作者信息
- 1. 北京交通大学 电子信息工程学院,北京 100044
- 2. 中国铁道学会,北京 100844
- 3. 北京交通大学 自动化与智能学院,北京 100044
- 折叠
摘要
列车的运行安全对轨道交通的可持续发展至关重要.牵引制动系统的故障可能导致严重的安全问题和运行延误.通过准确检测运行中的牵引制动故障,可以提高列车安全性,降低事故发生的可能性.本研究旨在提高列车运行牵引制动故障的诊断分类准确性与有效性.首先,基于时序特征的数据分析列车牵引制动故障数据集.设计数据重采样逻辑和特征提取方法,解决原始数据类别不平衡的问题,实现了显著的性能提升.实验表明,在使用相同的数据规模进行机器学习训练时,各个机器学习模型的分类性能均提升了10%以上.接着,建立基于SVM和决策树的机器学习模型,对数据集进行故障分类识别,并通过网格搜索进行超参数调整.此外,尝试多种基于基础分类算法的集成学习策略,并对所设计的模型进行仿真验证.其中,随机森林算法表现突出,测试准确率达到96.12%.通过分析分类报告和混淆矩阵,验证该方法的有效性和可靠性,为列车运行牵引制动故障诊断提供一种新的解决方案.
Abstract
Ensuring operational safety of trains is crucial for the sustainable development of rail transit systems,as faults in the traction and braking systems might lead to serious safety issues and operational delays.Precise fault detection can improve safety and diminish the likelihood of accidents.This study aims to improve the accuracy and efficacy of fault classification.Initially,this research conducted a data analysis based on the temporal features of train traction and braking fault datasets.A data resampling logic and feature extraction method were applied to address category imbalance in the raw data,enhancing data accuracy.Comparative experiments demonstrate that,with equivalent data volumes for machine learning and training,the classification performance of each model was improved by over 10%.Subsequently,machine learning models based on SVM and decision trees were established to identify and classify fault types,and the hyperparameter was optimized through grid search.Finally,various ensemble learning approaches based on basic classification algorithms were explored and the designed models were validated through simulation.Among them,the random forest algorithm emerged as the most effective with a test accuracy of 96.12%.Analysis of classification results and confusion matrices validates the effectiveness and reliability of this method,introducing an innovative solution for classifying fault types for train operation.
关键词
列车牵引制动/故障诊断/支持向量机/随机森林/集成学习Key words
train traction and braking/fault diagnosis/support vector machine/random forest/ensemble learning引用本文复制引用
基金项目
国家自然科学基金(62073027)
北京市自然科学基金(4232052)
北京市自然科学基金-丰台轨道交通前沿研究联合基金(L231017)
出版年
2024
NETL
NSTL
万方数据