Forecasting daily freight flow in cold regions of China using the hybrid Prophet model considering the importance of festivals and epidemic prevention policy

Abstract

Freight flow forecasting has emerged as a critical strategy for capturing future trends in freight variation and allocating resources effectively to ensure stable service for residents and local communities. Previous studies have primarily focused on forecasting annual regional freight demand and freight flow regarding freight trips or commodity flow, overlooking research into daily freight weight and parcels and the influence of various events on daily freight flow. This oversight neglects support for daily transportation tasks. In this study, we utilize a freight flow dataset comprising daily freight weight and parcels from a leading logistics company in cold regions of China. A hybrid XGBoost-SHAP and Prophet model is proposed to overcome the issue of Prophet failing to select important indicators, and to predict future freight flow and examine the correlation between freight flow and special events. Our findings reveal that the hybrid Prophet model outperforms LSTM, ARIMA, Prophet-SARIMA and the conventional Prophet model; meanwhile festivals and strict epidemic prevention policies have a significant impact on freight flow. These findings, derived from various contexts provincially, suggest that the Prophet model with events can serve multiple objectives in predicting freight flow, and contribute to the design of freight strategies.