基于地铁列车监控数据的牵引变流器滤网堵塞故障预警研究

Research on Fault Early Warning of Traction Converter Filter Fouling Based on Metro Train Monitoring Data

针对地铁列车牵引变流器冷却系统的滤网脏堵问题,提出了一种基于机器学习的变流器滤网脏堵预警模型,以指导滤网的预测维修决策。首先,为确定强分类的特征参数,结合列车牵引系统的检修记录,利用随机森林分类模型对列车历史数据进行特征筛选;其次,基于选定特征构建了孤立森林异常检测模型,并对模型参数进行优化;最后,分别在历史健康数据、故障数据集上进行测试,以确定模型能够识别滤网脏堵的准确度。结果表明:所构建的孤立森林模型分数能够直接反映牵引变流器滤网的脏堵程度,可有效指导滤网脏堵的维修决策。

Aiming at the problem of filter dirt and fouling of train traction converter cooling system,a converter filter fouling early warning model is proposed to guide the predictive maintenance decision of the filter screen.First,in order to determine the characteristic parameters of strong classification,combined with the maintenance records of the train traction system,the random forest classification model is used to complete the feature screening of the train historical data.Second,the isolated forest anomaly detection model is constructed based on selected features,and the model parameters are optimized.Final-ly,tests are performed on historical health data and fault data sets to make sure that the model can accurately identify filter fouling.Results show that the scoring of the constructed isolated forest model can directly reflect the degree of traction converter filter fouling,effectively guiding the maintenance decision for filter fouling.