Research On Movement Authority Generation Method Of Next Generation Train Control System Oriented To Tracking Trains In High Density

With the acceleration of urbanization in China,urban rail transit can meet the travel needs of passengers and alleviate urban traffic congestion to a certain extent.When the urban rail transit network and the line infrastructure have been determined,shortening the tracking distance between trains will be an effective way to further increase the traffic density and release the potential transportation capacity of the line.In the traditional train control system,movement authority(MA)is generated by the ground equipment.In the Next Generation Train Control System(NGTC),the on-board equipment of the following train can obtain the dynamic information of the preceding train in real time through train-train communication,and use artificial intelligence and other technologies to predict the dynamic behavior of the preceding train.On this basis,the on-board equipment can autonomously calculate MA and extend it to an appropriate position "over the preceding train".As a result,the following train can operate in the "soft wall" moving block,and higher-density dynamic tracking is achieved by shortening the tracking distance between trains.The main contents of the dissertation are as follows:(1)The theory of generating MA for NGTC is studied.First,the overall architecture of NGTC and the main functions of the on-board module are analyzed.Secondly,in high-density tracking scenarios,the MA generation methodology for NGTC is studied.And combined with the tracking operation modes of moving block,the principle of MA generation is proposed.Finally,the process of generating MA for NGTC is analyzed in detail.(2)The prediction models of train trajectory based on Bi-LSTM(Bi-directional Long Short-term Memory)neural network are constructed.With the train operation data from Sichuan University Jiang’an Campus to Wenxing Station on Chengdu Urban Rail Line 8,the Bi-LSTM method is used to establish the train dynamic time-space trajectory prediction models.In addition,three indicators including root mean square error,average absolute error and average absolute percentage error are used to evaluate the prediction models and the best hyperparameter combination determined by multiple experiments is applied to optimize the models.On this basis,the prediction results of the models are analyzed.It is found that when one-dimensional feature(running speed)is used for train speed prediction and two-dimensional feature(running speed and running distance)is used for train running distance prediction,the prediction accuracy of the models is the highest,and the obtained predicted trajectories fit the actual trajectories best.(3)A software for train trajectory prediction and tracking simulation is designed and developed.This software with functions such as train trajectory prediction,prediction result display,interval tracking simulation,and prediction model management is developed with C# language.And the tracking operation process of homogeneous trains in interval is simulated and analyzed.The results show that the dynamic distance between trains in the "soft wall" moving block based on the predicted trajectory of the preceding train is smaller than that in "hard wall".(4)The function of generating MA for NGTC is modeled and verified.Based on the theory of CPN(Colored Petri Net),three models of path resource management,associated train identification,and MA calculation are constructed.And through the analysis of the simulation execution results and the state space,the logical correctness of MA generation function is verified.Figures 74,Tables 23,References 85.