Prediction of delivery truck arrivals at container terminals: an ensemble deep learning model

Container terminal operators must balance external truck arrivals to the terminal and the prompt availability of yard resources. More accurate prediction of delivery truck arrivals is a crucial factor for the synergistic scheduling of yard operations. This paper proposes a novel ensemble deep learning approach to predict truck arrivals in a flexible period, with the span varying from one hour to twenty-four hours. With real data from the Yantian International Container Terminal in Southern China, multiple external nonlinear features are included in our deep learning model. Experiments demonstrate the effectiveness of the model which, among others, reveals the delivery pattern of shippers and truck arrival fluctuations. The decomposition of route-based prediction with cut-off time improves accuracy significantly. The results can be fed into the terminal operating system to improve the real-time scheduling of terminal operations. Furthermore, the announcement of predictions would allow customers to adjust their arrival time to avoid peak hours and this can be a good substitute or supplement to a truck appointment system.