An LSTM Approach for Predicting the Short-time Passenger Flow of Urban Bus

The advantages of LSTM as a deep learning neural network algorithm in time series prediction of passenger flow have gradually emerged. In addition to general passenger flow data that can be used for prediction, other context information can improve prediction performance. This paper proposed an LSTM approach for predicting the short-time passenger flow of urban bus, by using historical passenger data and other context information, e.g. weather type, holiday information and day of the week. The key parameters and structure of the long short-term memory (LSTM) neural network are deeply optimized. Adequate experiments with are conducted on the practical data of working day. The experimental result shows that the prediction of proposed LSTM outperforms the support vector regression (SVR) and k-nearest neighbor (KNN) algorithm. And the importing of weather data can improved performance of LSTM in the root mean squared error (RMSE) and the mean absolute percent error (MAPE).