Research On Fault Feature Extraction And Diagnosis Method For On-board Equipment Of CTCS Based On Labeled-LDA

With the continuous improvement of high-speed railway operation speed,as the core part of ensuring the safe operation of on-board equipment,the normal operation of the system is particularly important to ensure the safe operation of the train.In order to ensure the safe and efficient operation of the train,it is necessary to timely and accurately diagnose the types of faults that occur on the on-board equipment.In the current research of on-board equipment fault diagnosis,the diagnosis method is too dependent on the expert knowledge,and the fault diagnosis and analysis work is mainly dependent on the relevant technical personnel to do it manually.At present,this kind of analysis method cannot meet the needs of the current development of high-speed railway in our country.At present,the research of on-board equipment fault diagnosis is mostly based on the whole on-board equipment construction analysis model to complete the diagnosis of system failure.Such as the construction of Bayesian network or the establishment of on-board equipment fault tree model.However,there are few researches on the fault diagnosis of specific modules of on-board equipment,especially for the diagnosis and analysis of text data which records the state of on-board equipment.In this thesis,based on the log analysis of on-board equipment,Labeled-LDA method is used to extract the features of log data.Combined with particle swarm optimization support vector machine algorithm,the fault diagnosis of on-board equipment is realized.The main work of the thesisare as follows:(1)In this thesis,the characteristics of the fault data of the 300T on-board equipment are analyzed.According to the characteristics of fault data,the fault feature dictionary of on-board equipment is constructed by using the data mining method and combining the experience knowledge of the field technicians.Meanwhile,because of the one-sidedness of lexical feature extraction of text information,an improved Labeled-LDA is used to extract the semantic features of text information,and the text data is represented in the form of theme.(2)By statistics of fault sample data,it is found that the distribution of the fault sample data has a serious imbalance.In order to solve this problem,the particle swarm optimization algorithm is used to optimize the parameters of Support Vector Machine classifier to reduce the impact on unbalanced samples of the classification results.By comparing the simulation results,it can verify the effectiveness of the method in the classification diagnosis of unbalanced data.(3)Based on the above algorithm,the verification environment of on-board equipment fault diagnosis is designed.It realizes the feature extraction of the log data and the fault diagnosis of on-board equipment.In order to demonstrate the practicability of the verification environment,the fault data of six months in 2017 and 2016 were simulated.Compared with the actual fault records,the accuracy rate is over 90%.In this thesis,the fault log data of 300T on-board equipment is taken as research object.Combined with expert knowledge,a fault diagnosis method of on-board equipment based on improved theme model is proposed.The simulation results show that the improved Labeled-LDA is better than the Chi square statistics in the feature extraction of text data.Meanwhile,comparing with K Nearest Neighbor algorithm and Random Forest simulation results,it is verified that the method of particle swarm optimization to optimize the classification parameters of support vector machines can improve the classification performance of unbalanced data.