Research On Fault Diagnosis And Safety Assessment Of Train Control On-board Equipment Based On XGBoost Model

Railway has always been an important infrastructure of the country and a major artery of the national economy.In recent years,China’s high-speed railway has developed rapidly.During the Beijing Winter Olympics,the intelligent Fuxing EMU train sped along the Beijing-Zhangjiakou high-speed railway,this is a beautiful business card for China to showcase the high-quality development of railways to the world.The gradual opening and operation of high-speed railway lines have greatly improved the efficiency of transportation,and promoted the steady improvement of the economy.China Train Control System(CTCS)is an important component of the automatic control signal of the modern high-speed railway,which ensures the safe and efficient train operation.The on-board equipment of train control system is the core of the entire train control system.Whether the on-board equipment operates reliably will directly affect the safety of the train control system.Once the on-board equipment fails,it will interfere with the normal operation plan of the train and even endanger driving safety.At this stage,maintenance personnel mainly rely on expert knowledge and experience,and refer to the on-board AElog for fault diagnosis.Due to the complex structure of on-board equipment and unclear fault characteristics,misdiagnosis is prone to occur.Therefore,it is necessary to fully mine the text fault information,establish an intelligent diagnosis method,diagnose the fault type of on-board equipment timely and accurately,and realize the safety assessment of fault levels.The thesis takes the fault of CTCS3-300T vehicle-mounted equipment as the research object,uses natural language processing(NLP)technology to realize vector representation of text feature words,and uses the relevant theoretical basis of Ensemble learning to establish a fault diagnosis and evaluation model of on-board equipment based on XGBoost algorithm.The specific research contents are as follows:(1)Analyze the level of train operation control system,focus on the equipment composition and working principle of CTCS-3 train control system.Taking the most commonly used 300T on-board equipment as an example to analyze the specific functions of each module.Describing the source of original text fault data and the classification of fault categories.(2)The text fault data is preprocessed,and the fault feature thesaurus is selected after removing the deactivated words,building a professional knowledge thesaurus,and Chinese word segmentation.The improved TF-IDF method is used to calculate the weight of each fault feature word,and the feature word vector is discretized.The relevance of feature words to classification results is evaluated through importance ranking and SHAP values.The weight of the discrete word vector is taken as the result of feature extraction,that is,the input of the diagnostic model,and the ADASYN adaptive oversampling technology is used to synthesize the small class dataset.(3)Introduce the derivation process of integrated learning and XGBoost algorithm principles,and build the fault diagnosis model of on-board equipment based on XGBoost.Based on the text failure cases of Guangzhou Railway Group’s electrical section in recent three years,the Pycharm compiler is used to draw the confusion matrix,scatter plot and ROC curve after searching for the optimal parameter adjustment of the model grid.Compared with other machine learning SVM,RF and GBDT,the evaluation indexes of this model are significantly improved,with a diagnostic accuracy rate of 95.43%,verifying the superiority of this model.(4)The failure safety assessment level is divided based on the delay duration,and the failure safety results are evaluated by MAE,MSE,RMSE and R~2indicators,with the overall accuracy of 90.57%.Based on user’s demand analysis,a fault diagnosis and evaluation system for high-speed train control on-board equipment is designed and implemented by using Matlab and C#mixed programming.The system is divided into five modules,which respectively show the implementation process of each module.By establishing an XGBoost intelligent fault diagnosis model,it can improve the working efficiency of electrical maintenance personnel,reduce the occurrence of on-board equipment faults,and provide a reference for solving other textual fault diagnosis methods.