Fault Of Railway Signal Equipment Based On Text Mining Classification Research

Railway signal equipment is an important equipment of railway signal system and an important guarantee for the safety of railway traffic.In order to ensure the normal operation of the signal equipment,the railway system has established various safety monitoring/monitoring systems.These monitoring/monitoring systems generate a large amount of unstructured data stored in the form of voice,text,graphics and images.In daily maintenance and repair,the staff usually describes and record the causes of faults,fault phenomena,and fault handling methods in text form.Over time,a large amount of fault data is accumulated.These fault data are stored in the form of Word,Excel,etc.,mainly including related records when the accident occurred,post-tracing reports,driving logs,etc.,and finally archived in thesis form.These rich text data contain huge information,which is of great significance to the analysis of railway signal equipment.In order to promote the development of railway big data technology,the use of text mining technology to find out the connection between these fault texts and the occurrence of accident faults provides an informational and intelligent decision-making platform for railway traffic safety and realizes the intelligent classification of accident faults.This thesis proposes to use text mining technology to classify the text of railway accident fault text,and achieved the following results:(1)Use text mining technology to analyze the text of railway signal equipment accident fault.For the big data of railway signal equipment fault text,this article briefly introduces the signal equipment,expounds several commonly used fault diagnosis methods;and introduces the Chinese text processing method.According to the characteristics of the Chinese text,in order to achieve better word segmentation effect,jieba word segmentation technology is adopted,a professional dictionary in the field of railway is established,and the dictionary brought by jieba word segmentation is simply expanded.For the text feature representation,this thesis selects the Skip-gram model of Word2 vec to train the text after word segmentation to form the word vector representation of each word,and finally obtain the fault text word vector matrix,thereby improving the accident fault text feature representation effect.(2)Adopt an improved algorithm to deal with unbalanced text data.Since the types of faults are diverse,and the amount of data is different,the difference is relatively large,thus forming an imbalance between the category data.Aiming at the imbalanced text data in the fault,this thesis improves the SMOTE algorithm from the data point of view,and is used to process unbalanced data.The sample set is divided into regions,and samples in different regions adopt different sample processing methods.Finally,the original data,SMOTE algorithm,Borderline-SMOTE algorithm,SVM-SMOTE algorithm and the improved TSMOTE algorithm are compared,and it is concluded that the improved TSMOTE algorithm in this thesis can generate a better quality minority sample data set.Prepare for the breakdown text classification.(3)Study the classification model of convolutional neural network,and propose to use convolutional neural network model for text classification.With the popularity of deep learning in recent years,the convolutional neural network model is used to classify texts to solve the connection between word meanings often overlooked in machine learning,and the training process is prone to problems such as local optimization.In order to make the model have a deeper level of recognition ability,more comprehensive information can be extracted,so that the less important information in the sentence is not ignored,an attention mechanism is introduced,and the convolutional neural network model is optimized.Trained under the deep learning framework based on TensorFlow,the smaller Loss function value is obtained,which makes the accuracy of each category recognition reach a higher level.Finally,this article takes the actual accident fault text data of a railway bureau as an example,sets up multiple sets of comparative experiments from different angles,uses the proposed TSMOTE algorithm to deal with unbalanced data;uses the convolutional neural network model improved by the attention mechanism for grouping experiment.Compared with other classification models in the same field,it is concluded that the classification model proposed in this thesis has a significant improvement in classification accuracy,and it has also made a great breakthrough in the balanced data set,proving that the work done in this thesis This classification study has a certain contribution.At the same time,it also has a certain impact on the realization of railway accident fault analysis and the promotion of railway big data applications.