Research On The Application Of Extreme Value Theory In Safety Risk Analysis Of High-speed Railway

With the rapid development of China’s high-speed railway technology, the subject ofsecurity is more and more prominent in the high-speed railway operations management.Today, China’s high-speed railway operations have introduced the concept of riskmanagement. The study of risk analysis methods which are applicable to the current situationof China’s high-speed railway operations have become a hot topic.Railway accidents (especially high-speed railway accident) have low probability ofoccurrence, while once an accident occurs it will result in great losses. This feature is calledlow frequency-high loss characteristic. The curve of the data with this characteristic isshown as non-normal and heavy-tail.In this thesis, the author introduces the POT model of extreme value theory for riskanalysis of China railway accidents (especially high-speed railway accidents) which have thecharacteristics of this low-frequency and high-loss. This model can be a good solution to riskestimation and analysis of China high-speed railway accidents. In the case of selectingthreshold which is large enough, we fitting the data exceeding the threshold value in thesample to get the loss distribution function. And then we calculate the value at riskVaRPunder a given confidence level. Finally, we make an estimate of the loss of the accidentreferring to the previous calculation results.This thesis firstly introduces the development and the status of risk management at homeand abroad and some basic concepts, models and methods of risk management. The authoruses these concepts and methods into the risk analysis of China’s high-speed railwayoperation, and finds the proper process and model for China’s high-speed railway operationrisk analysis. Then the author uses the POT-GPD method to fit three different samples of data of somerailway operators in the history. The three different samples of data are the economic lossescaused by passenger train derailment accidents, the casualties caused by passenger trainderailment accidents and the economic losses caused by freight train derailment accidents.The author selects three different threshold values and uses the three sample of data to get theloss distribution function respectively. With the given confidence level P=95%and P=99%, the author calculates the value at riskVaR P. Then the author makes an analysisof the result of the value at risk of the given sample of data, and makes a test of the rationalityof the threshold value selection schemesFinally, this thesis focuses on the analysis of the threshold value selection and fittingresults of the loss distribution function and value at risk of the sample data. Then, the authorgives several suggestions for threshold value selection.