Abnormal Diagnosis Of Monitoring Data Of High Speed Railway Track Structure Based On Statistical Process Control And Sequence Probability Ratio Test

The high-speed railway has accumulated massive monitoring data in the process of operation monitoring in China.At present,the mining of monitoring data mostly focuses on basic statistical analysis such as data preprocessing,correlation analysis and prediction methods of monitoring data instead of using statistical time series method to analyze the monitoring data of track structure.Track structure monitoring data is a typical time series data,which has obvious discontinuity characteristics.In this paper,taking the monitoring data of track structure of a high-speed railway elevated station for three years as the research object,the statistical process control and sequential probability ratio test are introduced to analyze and mine the discontinuous track structure monitoring data by statistical time series anomaly diagnosis method.The main work and achievements are as follows:(1)Univariate statistical process control analysis:Multiple linear regression model and difference equation model were used to explain and separate common causes(temperature effect,linear trend and sequence dependence)in long-term monitoring data of track structure.Shewhart control chart and EWMA control chart were used to diagnose the data changes caused by special cause for the residual sequence after eliminating the influence of common causes.The results showed that besides temperature effect,linear trend and sequence dependence were also important sources of monitoring data changes caused by common causes.For each monitoring point,multiple control charts simultaneously diagnosed two kinds of most prominent special events.For multiple monitoring points,multiple control charts simultaneously diagnosed three types of prominent special events.(2)Multivariate statistical process control analysis:For the residual sequence after eliminating the influence of common causes,T~2 control chart and MEWMA control chart were used to analyze typical cases of monitoring data sequence.It was found that extreme temperature change was an important reason for the abnormal value of T~2 control chart.The source of outliers in T~2 control chart can be analyzed by drawing elliptical control area.The combination of T~2 control chart and MEWMA control chart can diagnose whether the abnormal value has a lasting effect greater than 1 day.The two most prominent special events in the monitoring process were obtained based on the outliers diagnosed by multivariate control charts.(3)Sequential probability ratio test analysis:Multivariate state estimation technology and difference equation model were used to model the monitoring data in the training stage,then the monitoring data in the inspection stage were fitted to obtain the model residue,and finally the sequential probability ratio test(SPRT)was used to diagnose the abnormal model residue.It was found that multivariate state estimation technique(MSET)had good prediction effect.The difference equation model can effectively eliminate the autocorrelation in MSET residuals.The number of outliers diagnosed by SPRT can be used to judge the state of track structure.The most prominent special events in the monitoring process were obtained based on the abnormal values diagnosed by SPRT results of multiple monitoring points at the same time.There are 70 figures,24 tables and 78 references.