Passenger flow forecast of railway station based on improved LSTM

Abstract

To solve the problem that the traditional neural network model lacks the ability to predict the complex nonlinear data, this paper constructed a short-term passenger flow prediction model based on the improved LSTM. Taking the AFC data of Beijing West Railway Station as the research object, the neural network model is trained by using the deep learning framework Keras. The prediction results of the improved LSTM network model is compared with BP network model and the standard LSTM network model. The results show that the improved LSTM model has better prediction results. In different periods of weekdays and weekends, the mean absolute percentage error (MAPE) of passenger flow prediction is lower than other models.