Prediction of Short-term Passenger Flow in the Metro Station with CNN-LSTM Model

Nowadays, the overwhelming majority of people choose to go to work during commuting time, causing great pressure on the traffic system. And methods of effectively predicting the short-term passenger flow of each subway station have aroused great concern. Traditionally, the LSTM deep learning model will be commonly used to process time series data. The reason is that it not only tackles the issue of long-distance reliance that RNN cannot deal with but also solves the trouble such as gradient explosion or gradient disappearance that are common in neural networks. However, in comparison to the LSTM model, the CNN-LSTM model makes it easier to extract the characteristics of time and space. Hence, it is necessary to evaluate which model has a better capacity for predicting passenger flow in metro stations. Undoubtedly, it can find that the CNN-LSTM model can accurately predict the daily traveler stream due to the excellent space extraction ability, through the evaluating function of MAE, MSE, and MASE. Additionally, the development forecast of passenger flow of urban transportation system is a basic work for the construction of urban transportation facilities. Therefore, its role is particularly prominent in the planning and construction of urban rail transit.