The HWAM-EMD-GRU Forecasting Model for Short-Term Passenger Flow in an Airport Light Rail Transit Line

Abstract

Abstract Accurate forecasting of airport light rail transit line (ALRTL) outbound passenger flow is critical to the optimal operations of both light rail and airport systems. Considering the nonlinearity, non-stationarity, uncertainty, and periodicity of outbound passenger flow in the ALRTL, we propose a combined forecasting model that integrates the Holt and Winters additive model (HWAM), empirical mode decomposition (EMD) and gated recurrent unit (GRU). Firstly, the edge effect of EMD will greatly affect the performance of the forecasting model. To overcome this, we extend the passenger flow by HWAM. After that, the decomposition method, EMD, can be applied to passenger flow, and several intrinsic mode function (IMF) components can be extracted. After extracting all the IMFs, the remaining part is referred to as the residual (Res) component. Then, a correlation test is performed on all the components, followed by their aggregation. Finally, the GRU is used to predict each of the aggregated components, and the prediction of aggregated components requires reconstruction. To verify the performance of the HWAM-EMD-GRU, we conducted a comparative study on the hourly passenger flow data for Beijing Daxing International Airport Express and set the autoregressive integrated moving average model, HWAM, Prophet, and GRU as the baseline. Predictions of the HWAM-EMD-GRU combined model demonstrated higher accuracy than baseline models, with a root mean square error of 83.52 (Prophet is 110.21) and mean absolute percentage error of 8.32% (Prophet is 12.48 %). The experimental result shows that the HWAM-EMD-GRU forecasting model offers more accurate predictions.

Keywords

• Passenger flow forecasting
• Airport light rail transit line (ALRTL)
• Empirical mode decomposition (EMD)
• Gated recurrent unit (GRU)
• Holt–Winters additive model (HWAM)