End-Edge-Cloud Collaborative Learning-aided Prediction For High-speed Train Operation Using Deep Learning

Train operation control system is an important facility to realize operation command and over-speed protection in high-speed train operation process,which aims to ensure the safe and smooth running of trains.While the information available to the train driver is limited from the human-computer interaction interface of train operation control system in high-speed train operation process,which causes that the internal and external unknown disturbance are dealt ineffectively.At the same time,due to the complex and changeable operation environment of high-speed trains,the train operation is greatly affected by the personal ability and technical level of the driver,which will affect the safety and punctuality of train operation smoothly.Therefore,in the operation process of high-speed train,it is of great practical significance to design a method that can effectively realize the accurate prediction of the traction/braking control process of high-speed train,for realizing the efficient control of the traction/braking control unit of high-speed train.Aiming at the problems existing in the modeling of the operation process of high-speed trains,a model oriented to edge computing for HST is proposed based on the analysis of the current status of high-speed train driving and control and the shortcomings of the current control mode,with the CRH380 B train operation process as the research object.At the same time,in view of the massive and diverse data and the difficulty of centralized processing,an end-edge-cloud collaborative architecture is proposed to apply to high-speed railway and the end-edge-cloud collaborative learning-aided framework for HST operation process is established in this paper.And then deep neural network in deep learning technology is used to complete the training of high-speed train operation process prediction model,to realize the combination of dynamics and data driving and the construction of the end-edge-cloud collaborative learning-aided prediction model for high-speed train operation process based on deep learning.And different cases are set to verify the prediction performance of the model for HST operation process.The specific research contents are as follows:(1)Considering the large amount and variety of the data in high-speed train,it is difficult to deal these with the traditional centralized analysis method.This paper develops a model oriented to edge computing for high-speed train,according to the dynamic characteristics and information flow direction of train data,as well as the working characteristics and functions of cloud computing and edge computing.In addition,the end layer,edge layer and cloud layer of high-speed train are correctly divided,the the functions of the end layer,edge layer and cloud layer are determined and the data information flow between the end layer,edge layer and cloud layer are clearly defined,so as to build an End-Edge-Cloud collaborative learning-aided framework for HST operation process.(2)With the collected rail and train field data,a HST operation data preprocessing method for model training in deep learning is designed.Then with the base of the analysis of the working principle and basic structure of recurrent neural network,combined with the end-edge-cloud collaborative learning-aided system of high-speed train operation process proposed in(1),an end-edge-cloud collaborative learning-aided prediction method of high-speed train control process using the recurrent neural network is proposed.Based on the preprocessed data,different cases are given to train and test the prediction performance of the proposed method on the running speed and traction/braking handle level prediction in high-speed train operation process.(3)Aiming at the problems existing in the end-edge-cloud collaborative learning-aided prediction method using recurrent neural network of(2),a prediction method of end-edge-cloud collaborative learning-aided prediction for HST operation process using long short-term memory neural network is developed with the combination of deep learning technology and end-edge-cloud technology.In last,different cases are given to train and test the prediction performance of the proposed method on running speed and traction/brake handle level in high-speed operation process.