基于机器学习的受电弓滑板磨耗预测模型

Wear Prediction Model of Pantograph Contact Strip Based on Machine Learning

为研究受电弓滑板摩擦磨损性能的影响因素,根据磨耗演变规律对磨耗进行预测,采用置信区间估计法,确定滑板历史磨耗数据统计值上下界和基准训练集,建立机器学习的线性回归模型,以梯度下降法使代价函数趋于最小对模型进行优化。通过对该模型及方法的应用,预测滑板剩余厚度限集,并通过与某型车实测磨耗数据比较。结果表明:预测数据与实测基本一致,可为有效减少动车段对受电弓滑板维护工作量提供依据。

In order to study the factors affecting the friction and wear performance of pantograph strip,the wear of the strip is predicted in light of wear evolution law.The confidence interval estimation method is used to determine the upper and lower bounds of the statistical value of the historical wear data of pantograph strip and the benchmark training set,based on which,a linear regression model of machine learning is established,and gradient descent method is applied to minimize the cost function for the optimal results of the model.With the application of the model and the method,the residual thickness limit of the strip is predicted,and its results are compared with the measured wear data of a certain type of train.The results show that the predicted data are basically consistent with the measured one,which can provide a basis for reducing the maintenance workload of the pantograph strip in motor car depot.