Comparison of NARNN and ARIMA Models for Short-Term Metro Passenger Flow Forecasting

Accurate short-term forecast of real-time metro passenger flow is essential for improving the operation and management level of a metro system. This research establishes the linear model ARIMA (autoregressive integrated moving average) and nonlinear model NARNN (nonlinear autoregression neural network) respectively to forecast the short-term entrance passenger flow of a metro station. The application experiment is based on AFC data of Confucius Temple station of Nanjing metro line three. Prediction time interval is set as 5 and 15 min. Comparison results show that accuracy of both models is higher in 15 min interval than in 5 min interval. NARNN has greater accuracy and adaptability than the ARMIA model. However, there is slight lag in the forecast of the NARNN model, while the prediction errors of the ARIMA model increase with time. The results indicate a requirement for optimal or combined methods to improve the accuracy of shorter time interval forecasting.