The Forecasting of Train Occupancy Rate on High-Speed Railway Based on Long Short-Term Memory

The train occupancy rate (TOR) is an important indicator to measure the matching relationship between supply and demand of high-speed railway passenger transport. The railway passenger transport department usually evaluates the actual driving effect of the train diagram by TOR. Before the actual timetable is executed, the TOR is unknown and it is difficult to describe it with an accurate matching model. In recent years, with the rapid development of China's high-speed railway, the available operational data has increased dramatically, which has prompted us to no longer use traditional applied mathematics methods, but to use deep learning methods to improve the accuracy of predicting TOR. In this paper, long short-term memory (LSTM) is used to predict TOR. The specific methods of feature extraction and data input are discussed. Considering the variation law of passengers under multi-factors, the operation data of Beijing-Shanghai high-speed rail between 2012 and 2018 is utilized to verify the effectiveness of this model. The final experiments show that the results obtained in this paper are better.