CIR Device Failure Analysis And Prediction Based On Semantic Analysis

With the continuous improvement of the degree of informationization of railways and the increasing demand for multi-service integration,railway researchers need to quickly obtain valuable information from complicated business data.The traditional manual methods and traditional data processing tools cannot match the demands of massive data calculation analysis and real-time monitoring of train operations.Cab Integrated Radio Communication Equipment(CIR in short)is an important device to ensure train dispatching and control,and to ensure railway safety.It processes important information such as train dispatching communication,alarming,voice communication and train rear wind pressure query.The CIR equipment records a large amount of equipment operation data during the train operation,which directly reflects the operation status of the equipment.This thesis attempts to process and analyze the CIR data to achieve fault analysis and real-time prediction of on-board equipment,and provide constructive advice for CIR equipment maintenance and upgrade.Through data analysis,it is possible to perform preventive,in-process diagnosis and fault positioning,and post-evaluation functions on equipment failures,thereby providing data support for decision-making of equipment maintenance work.This thesis conducts in-depth analysis and research on CIR data.Based on its similarity to natural language characteristics,we use deep data mining models such as semantic analysis and natural language learning models to conduct in-depth analysis.We bring improvement to infinite hidden Markov model and n-GRAM language model.The improved Beam sampling infinite hidden Markov model and LSTM-n-GRAM model are proposed,which realizes the CIR data state classification and fault real-time monitoring.Compared with the traditional manual data analysis method,it has the feature of larger data and higher accuracy,faster query and stronger real-time performance.The implementation of these two models is different from the theoretical basis,but the conclusions reflect that the CIR data has Markovity,enriching the theory of train failure analysis and real-time prediction,providing solutions for equipment maintenance and fault diagnosis,and railway transportation.The actual application and verification have been obtained in production,which proves that it has theoretical value and engineering application value.The main inovative research is as follows:Firstly,based on the time-series characteristics of CIR data,combined with the Markov assumption,the improved infinite hidden Markov model-Beam sampling infinite hidden Markov model is used to classify CIR data,and the normal state and the unknown number of abnormal states are performed.The division realizes the effective monitoring of the CIR state and verifies it by combining the actual train running CIR data.Secondly,considering the real-time requirements of CIR fault monitoring,we use the improved n-GRAM model-LSTM-n-GRAM model for state prediction,and using CIR data to get the actual verification.Experiments show that the short-term data using n-GRAM is not significantly different from the performance of the model using the entire time series data.The LSTM-n-GRAM model does not need to use all the data for data analysis,which greatly improves the calculation efficiency and avoids gradient disappears and other issues.Thereby greatly improving the real-time performance of CIR condition monitoring.