Research And Application Of Track Dynamic Inspection Data Mining Algorithm

Track dynamic inspection data is a multivariate time series data reflecting the health status of the track.Mining and analyzing it will help to study the changes of the track status,provide precise guidance for further maintenance and repair,and at the same time can be in-depth exploration the causes and development trends of railroad diseases,timely warning,and improve the safety of line operations.Multivariate time series is a key research direction in the field of data mining in recent years and is widely used in many industries.It is of great theoretical significance and practical application value to study the analysis and processing methods of multivariate time series and to mine the characteristics of track dynamic inspection data.Based on the theory of multivariate time series data processing and the track dynamic inspection data of a high-speed railway in past 8 years,this paper studies the mining algorithm of track dynamic inspection data,focusing on the identification of track differential settlement and the abnormal detection of track dynamic inspection data:(1)An identification method of track difference settlement based on Siamese Neural Network is proposed.This method starts from the similarity measurement of the track dynamic inspection data sequence,and uses the Siamese Neural Network to learn the similarity measurement,and builds a differential settlement identification model of the Siamese Neural Network based on LSTM.The experimental results on the track differential settlement data set composed of real data of a high-speed railway line shows that the precision rate of the differential settlement identification of the model is 96.23%,the error rate is 3.77%,and no false alarms have occurred.This method provides a new idea for the correlation research and analysis of track dynamic inspection data and track differential settlement data,and also provides technical support for the identification and early warning of track settlement status.(2)An anomaly detection method for multivariate track dynamic inspection data based on the autoencoder is proposed.This method deeply studies the multivariable correlation relationship in the track dynamic inspection data,constructs the multivariable correlation matrix,and proposes a multivariate track dynamic inspection data anomaly detection algorithm based on the autoencoder(MTDAD-AE).The experimental results on the real track dynamic inspection data set of a high-speed rail line show that the proposed algorithm is superior to other deep learning-based comparison methods in precision and recall rate,and has achieved the best detection results.In addition,based on the proposed correlation matrix,the cause of anomaly can also be located and analyzed.(3)An intelligent processing platform for track dynamic inspection data(IPP-TDID)was constructed.The platform adopts bottom-up design ideas,including data acquisition layer,data processing layer,data analysis layer and data application layer.The platform uses Spring Boot and Vue to realize the visual display of back-end data business logic and front-end pages.It integrates intelligent algorithms for track dynamic inspection data,including mileage offset registration algorithm for track dynamic inspection data based on correlation coefficients,track differential settlement identification model based on siamese neural network,and anomaly detection model for track dynamic inspection data based on auto-encoder.The construction of the platform supports the early scientific research phases such as data cleaning,which greatly speeds up the scientific research process of track dynamic inspection data.At the same time,the integration of intelligent algorithms provides the basis for the continuous research of track dynamic inspection data mining and application in actual line operation scenarios.It has important practical application value.