Research On Integration Reduction Of High-speed Railway Catenary Data Based On Feature Selection

As an important part of traction power supply system of high-speed railway,catenary is arranged along the line.The catenary is easily affected by meteorological factors such as rain,snow,gale,sand and dust.In the process of pantograph-catenary operation,it is easy to cause defects such as abnormal geometric parameters,loosening and breaking of parts,which affect the safety and stability of traction power supply system.Therefore,it is necessary to detect and monitor the status information of the catenary,and to maintain the catenary fault intelligently after analyzing the monitoring data.The original data information collected has many characteristics,such as data types,including numerical data representing geometric parameters,narrative sentences representing parts status,lack of consistency in expression,difficulty in comparative analysis of data,which makes large data streams not fully utilized.There is redundancy among data,and too high data dimension leads to too high cost of time and space for learning analysis,even can not be analyzed directly.In order to solve the above problems,this paper designs a rigorous and standardized catenary data dictionary for high-speed railway,which makes the original complex and disordered data become standardized and orderly data,and integrates and reduces the redundancy of data through feature selection algorithm,reduces the data dimension,facilitates subsequent data mining and acquires valuable knowledge.Based on the analysis of the related specifications of the catenary,the basic framework of the catenary data dictionary for high-speed rail is proposed based on the data dictionary and the multi-dimensional structure model.The data dictionary system of high-speed rail catenary under this model is established,and the items are classified and defined.In order to standardize the expression of catenary data dictionary in high-speed railway,the principles of attribute expression at each level are determined under multi-level structure,and the related attributes are coded informationally.The expression form of data dictionary of high-speed rail catenary is determined,and its coding format and parameter setting are explained in detail.Based on the data dictionary,data identification and other integration technologies are used to standardize the content and format of catenary data,which lays a data foundation for subsequent analysis and research.Based on feature selection algorithm,the fault data of catenary of high-speed railway are integrated and specified.The existing data structure and feature selection algorithm are deeply studied.Because the catenary fault data of high-speed railway are labeled-free data,this paper firstly uses AGNES hierarchical clustering method to cluster catenary sample data and assign corresponding labels.Then Relief F algorithm is taken as the core of this paper according to the performance of each feature selection algorithm.Aiming at the deficiencies of Relief F algorithm,such as redundancy,repetitive sampling,etc.On the basis of ensuring random sampling and based on the idea of stratified sampling,a kind of sample is regarded as a stratified one,so as to control the number of times that all kinds of samples are selected in the overall sampling;The Jacquard coefficient is introduced into the commonly used Pearson similarity calculation method as a correction factor.Then the redundancy features are evaluated and removed.Therefore,a feature selection method suitable for unsupervised catenary fault data is proposed and applied to the actual catenary fault data of high-speed railway.The validity of this method for catenary data is verified by comparing with SPEC algorithm.