Intelligent Prediction Of Train Delay Changes And Propagation In High Speed Railways

With the construction of China’s High-Speed Railway(HSR)and the density of trains increasing,when some trains are delayed due to the interference of emergencies in the network,it is necessary to limit speed of relevant trains or adjust the arrival and departure time of trains to ensure the safety and transportation service quality.The above measures will cause the delay of train.If the delay is not eliminated in time,it will lead to the spread of delay to subsequent related trains,resulting in more trains’ joint delay.Due to the different actual situation of each station and running section,the utilization rate of redundant time of train diagram is different,which makes it difficult to predict the change of train delay time.However,the time-varying and strong coupling of train interaction makes it difficult to describe the train delay propagation model and predict the train joint delay based on the existing rules.Therefore,the establishment of data-driven prediction model of high-speed railway train delay based on train performance data is a supplement to the existing basic theoretical research of high-speed railway,which can provide model support for the research and verification of intelligent scheduling algorithm,and help dispatchers to judge the train operation situation and make real-time scheduling decision,It is of great significance to improve the quality of railway traffic command and dispatching,and to optimize the train timetable and adjust the structure of train diagram.Based on the National Natural Science Foundation major project "Basic Theory and Key Technology of Integration of High-Speed Railway Operation Control and Dynamic Dispatching"(61790574),this thesis analyzes and predicts the delay time variation and delay propagation of high-speed trains in the operation process under the background of high-speed railway operation dispatching under the jurisdiction of China Railway Shenyang Bureau.The specific work is summarized as follows:(1)The overall structure of delay change and propagation prediction model is proposed.The problem is divided into several parts:regression prediction of initial delay change of train,classification model of joint delay judgment of train,regression prediction of associated delay duration of train,and error compensation model of train delay prediction The judgment of the scope of communication.(2)The RVFLNs algorithm based on stacking ensemble learning(SRN)is proposed as the regression algorithm.Based on the basic RVFLNs,the SRN algorithm takes the input of weak learners and the output of weak learners as the model input of the combination strategy part.The final learner combined with the strategy part is trained to complete the training of the model.It solves the problem of recovery prediction in the change of high-speed railway delay,and the prediction problem of joint delay time after the initial delay occurs.The prediction model of delay change and the input and output variables associated with the delay time are explained respectively,and the correlation analysis and simulation experiment with actual data are completed.The distribution of time length under different causes of possible secondary delay is fitted.(3)In this thesis,we propose an improved algorithm for data migration based on the improved SMOTE algorithm(ITRN).It solves the propagation classification judgment of high-speed railway delay.Due to the lack of marked data in some sections and stations,it is impossible to establish an effective model.RVFLNs classification algorithm(TRN)based on transfer learning is proposed by introducing the migration learning algorithm.Moreover,there is a large imbalance between labeled data and unlabeled data.An improved smote algorithm is proposed based on TRN algorithm to form the final ITRN algorithm.The input and output variables of the delay propagation classification algorithm are analyzed,and the correlation analysis and the simulation experiment based on the actual data are completed.(4)For the error of delay prediction modeling,due to the inevitable error of the prediction model,it is bound to cause error accumulation in the iterative delay propagation chain.In order to reduce the prediction error,the new modeling variables including the distance between the train and the distance between the train and the terminal station are added to further reduce the modeling error.The effectiveness of the new variables is proved by analyzing the error distribution curve.According to the idea of stack model of ensemble learning,the second layer error compensation model is established according to the prediction error of delayed regression.In this thesis,through the establishment of stacking integrated learning model and the introduction of transfer learning and improved SMOTE oversampling algorithm classification model,the change and propagation process of high-speed train delay in the process of operation are modeled and predicted,and the change of high-speed train delay time and continuous belt delay are predicted,which solves the data imbalance in the above process The prediction accuracy of the model is low.