摘要
以浸金属碳滑板、铜导线为摩擦副进行对摩实验,对弓网系统接触失效问题进行研究.建立基于载流相对稳定系数、载流效率、离线率的弓网接触失效判定标准,依此将接触状况分为失效、有效两种状态.进而采用随机森林(RF)和SVM算法分别建立接触压力、压力波动幅度、压力波动频率、滑动速度和接触电流与接触状况的二分类模型,并利用交叉验证方法比较两种模型的性能.依据RF和SVM-RFE算法共同分析了各工况变量对接触失效的影响程度,并分析得到接触失效率随各工况变量的变化趋势.算例表明:RF模型的精度和AUC值分别为92.8%和0.883,并且模型具有良好的预测稳定性;接触压力和滑动速度对接触失效影响较大,二者重要程度为总体的67%;且失效率随着接触压力的增大呈下降趋势,随滑动速度提高而上升.
The contact failure of pantograph catenary system is studied by friction test with the Cu impregnation carbon material and Cu wire.Firstly,the contact failure judgment of pantograph catenary system are established based on the current stability coefficient,current carrying efficiency and off-line rate.According to these criteria,the contact condition is divided into failure/effective states.The two classification models of contact pressure,pressure fluctuation amplitude,pressure fluctuation frequency,speed,current and contact condition are established by using random forests(RF)and SVM algorithm,and performance of the two models is evaluated by cross validation method.Then,according to the RF model and SVM-RFE algorithm,the importance of each variable to contact failure is analyzed.Finally,the change trend of contact failure rate with each variable is analyzed.The results show that the accuracy and AUC value of RF model are 92.8%and 0.883 respectively,and the model has good prediction stability;the importance of contact pressure and sliding speed to contact failure is 67%of the total;and the contact failure rate decreases with the increase of contact pressure and increases with the increase of speed.
作者
于淼
回立川
陈忠华
YU Miao;HUI Lichuan;CHEN Zhonghua(Faculty of Electrical and Control Engineering,Liaoning Technical University,Huludao 125105,China;State Grid Huludao Electric Power Supply Company,Huludao 125000,China)
出处
《辽宁工程技术大学学报（自然科学版）》
CAS
北大核心
2022年第5期462-469,共8页
Journal of Liaoning Technical University (Natural Science)
基金
国家自然科学基金(51477071)
辽宁省教育厅科学研究项目(LJ2017QL009)
辽宁工程技术大学生产技术问题创新研究基金(20160021T)
关键词
弓网系统
滑动电接触
接触失效
随机森林
失效预测
pantograph-catenary system
sliding electric contact
contact failure
random forests
failure prediction
