Pantograph-Catenary Fault Diagnosis Based On Data Mining

With the rapid development of transportation industry,the country attaches great importance to the construction of railway.Electrified railway is becoming more and more popular with rapid,stable and clean energy as its development direction.As the main power supply and receiving equipment of electrified railway,pantograph catenary is more prone to problems as the speed of high-speed trains increases.To ensure the safety of the train,avoid the failure of the pantograph network,find out the hidden danger from the data parameters of the high-speed train detection,some data in the pantograph are independent,some data are related,and every data may be in accordance with the requirements,but the combination of them may be hidden.This paper realizes the diagnosis of pantograph catenary fault by studying the inherent relationship between data parameters.In this paper,the rough set method is used to analyze the operation parameters of the pantograph network,and the rough set can be used to deal with the inaccurate and fuzzy data.At the same time,the data parameters of each data can not be very complete.The hard points,impact,contact wire height,contact pressure can be obtained without affecting the important data properties.The data of 10 kinds of parameters,such as force,anchorage joint and pull out value,are analyzed,and 6 kinds of fault types are studied,such as suspension clue fault,basic fault,wrist arm fault,contact wire fault,drill bow fault and positioning device fault,find out the rules in the middle,establish expert system,judge and analyze the cause of pantograph network fault.Reduce the occurrence of accidents..Before the simplification of rough sets,the data need to be preprocessed.Many data mining results are not ideal because the quality of the data is not high.Because of the influence of environment and other factors,the collected data must contain a large number of errors and noise.Data preprocessing is the necessary integration,cleaning,dispersion and reduction of the original data before the analysis and mining of the data,and then using the Adaboost-SVM algorithm,the algorithm can effectively combine the advantages of the two and make up a part of the shortcomings,and then improve the accuracy of the classification of the whole body.