Study On The Primary Delay Affect Model By Using The Real Operation Data Of Guangzhou Railway Bureau High-speed Railway

High-speed railway(HSR)trains will be inevitably disturbed by internal and external factors during the operation process.A large disturbance,if it cannot be assimilated by time supplement or buffer in the timetable,could cause train delays,which is caled primary delay.Due to the appropriability of railway resources,the primary delay may propagate in time and space,which is caled knock-on delay or secondary delay.Therefore,uncovering the influencing indexes of primary delay,fitting primary delay length,establishing station-level predicting models,and completing the accurate prediction of train delay can provide reliable decision supports for real-time train dispatching and improve the quality of transportation service.As for fitting the distribution model of primary delay influencing indexes,in this thesis,we divided the railway network controlled by Guangzhou Railway Bureau(GRB)into three levels,which are HSR lines,Wuhan-Guangzhou HSR line and segments of Wuhan-Guangzhou HSR line.Then,distribution models were established,using the railway disturbance records in 2014 and 2015,and validated using the disturbance records in 2016.The accomplished tasks include:(1)the primary delay distribution model with the best goodness-of-fit was obtained using the Kolmogorov-Smirnov test in the three spatial levels respectively;(2)the number of affected trains and total delay times of primary delay were fitted using distribution models and selected according to their goodness-of-fit based on Kolmogorov-Smirnov test and R~2 coefficient in the three spatial levels respectively;(3)the applicability of the number of affected trains and the total times of primary delay distribution models and was verified using the disturbance records data in 2016,based on matching degree.For prediction of station-level affected trains and total delay times of primary delay,the delayed train sequences of GuangzhouNorth and HengyangEast station were extracted based on the train operation records during 2015-2017 in GRB.Prediction model for the number of affected trains was established,using the historical train operation data of GuangzhouNorth station real data in 2015 and 2016.The historical train operation records of Hengyang East station was used to validate the applicability of the model in the spatial dimension,and the Guangzhou North Station data in 2017was used to validate the applicability of the model in the temporal dimension.The result indicated that XGBOOST algorithm had the best goodness-of-fit on training dataset and testing dataset.Then,based on the predicting models for affected trains,predicting models for total delay times of initial delays was established using the train operation records of GuangzhouNorth station in 2015 and 2016.Also,the established model was validated using the historical train operation records of HengyangEast station in 2015and 2016 and records of GuangzhouNorth station in 2017.Finally,the model training and testing process showed that support vector regression(SVR)had the optimal predicting accuracy.