High Speed Railway Delay Forecasting Method Based On Artificial Neural Network

Nowadays with the expansion of China High Speed Railway(CHSR)network and the improvement of the service quality of railway travelling,High-speed train has become a principle travel mode in China.By 2025,China's railway network is expected to reach 175,000 kilometers,with high-speed railways accounting for more than 20 percent of the total length and covering more than 80 percent of major cities.The characteristics of High speed and high density in high speed railway operation mode bring tremendous challenges to the railway transportation organization.High-speed trains are inevitably disturbed by various factors in the process of operation,leading to train delay,which involves not just the operation of this train,but also propagates throughout the network,causing even more serious delay situation.So,it would be significant for the spot dispatching to monitor the train delay situation and forecasting the train delay in real time.Since delay levels enables dispatchers to catch the delay information more sensitively,a train delay level forecasting model based on Artificial Neural Network(ANN)was built in this dissertation.The major researches are:(1)The mechanism of forecasting train delay with ANN was analyzed,and the mode of using the delay situation of the prior trains to deduct the delay of trains behind was established,according to which the form of the historical delay sample was confirmed,and the classification of train delay levels was given.(2)To train the ANN model and to meet the need of model,historical delay data were selected and filtered from the database,with considering the feature of train graph,the train delay sample set was built.(3)Based on the data processing in Deep Learning,the train delay sample set was normalized,regularized and balanced.(4)An ANN model was built for data training,and the parameters were adjusted repeatedly through backward propagation to diminish the difference between the output of the model and the target value.(5)The structure and the algorithm of the model were optimized.To confirm the reliability of the model,comparisons were made with human forecasting and Multiple Linear Regression Model.The training result of the two ANN models in this dissertation shows that,ANN model can effectively fit the influence of the prior trains on the rear trains and can forecast the rear train delay levels.The application scenes of train level forecasting in the dispatching system were simulated,showing how the model serves the dispatchers to improve work efficiency.