Research On Forecasting Method Of Wheel Wear Based On Data

With the rapid development of modern detection technology and modernrailway industry automation technology, a large amount of measured data of wheelsystem was saved by China railway related department. It contains rich characteristicsand laws information of wheel rolling system. In addition to the large volume of thesedata, they also have the features of update speed, nonlinear, high correlation, and theyare usually stored in a high dimensional space by unstructured form. We have to use avariety of data analysis and reduction method to reveal the nature of these data in highdimensional. This paper studies the prediction method of wheel wear volume basedon manifold learning.This paper explores the application area of manifold learning algorithm;Themanifold learning algorithms are applied to forecast model and the modeling of thewheel wear system. Firstly, to deal with the laminations of the unknown sample dataprediction based on the traditional WDMR model, a new model of weightdetermination by manifold regularization (WDMR) is proposed. Then, consideringthe integrity and relevance of each train wheel wear volume, the prediction model ofmulti wheel wear volume based on LLE algorithm is proposed. Lastly, in order toimprove the performance of the model, a new prediction model of LLE-WDMRoptimized by CSA is proposed, the simulation results show that the model has betterfitting and stable performances. Then, prediction of Wear Volumes of Wheel/Railbased on Subspace Identification Method is proposed, and the model has betterperformances.