Passenger Flow Prediction Based on a Hybrid Method in the Nanjing Metro System

Accurate short-term passenger flow prediction provides key information for operation and management of urban rail transit systems, which is becoming more significant in this field. In this paper, a novel hybrid model DWT-SARIMA is proposed, and it combines two models of discrete wavelet transform (DWT) and seasonal autoregressive integrated moving average (SARIMA). The proposed model includes three critical stages. The first stage decomposes the passenger flow data into different high frequency and low frequency series by discrete wavelet. In the prediction stage, the SARIMA methods are applied to predict new high frequency and low frequency series. In the last stage, the predicted sequences are reconstructed by discrete wavelet. The experiment results showed that the proposed model had the best forecasting performance compared with SARIMA and a back-propagation neural network (BPNN) model, and it had a more reliable prediction results [mean absolute percentage error (MAPE), variance of absolute percentage error (VAPE), and root mean square error (RMSE) were 9.372%, 0.670%, and 36.364, respectively] based on the No. 2 metro line in Nanjing metro system.