Ship arrival prediction and its value on daily container terminal operation

Abstract Effective prediction of ship arrivals should provide the estimated delay or advance of arrival ships with greater accuracy, and improve the performance of container terminal operations. Therefore, taking Gangji (Yining) Container Terminal (GYCT), China, as an example, this paper resorts to data mining approaches to predict ship arrivals and explore the value of such predicted ship arrivals on the container terminal operation. First, this study applies three data mining approaches, including Back-Propagation network (BP), Classification and Regression Tree (CART) and Random Forest (RF), to estimate the delay or advance of ship arrivals using the collected data of ship arrivals. Then the predictive performance of these three approaches is compared and discussed, it is concluded that RF performs better than BP and CART, and ETA month and ship length are the most important determinants of ship arrivals in GYCT terminal. Finally, series simulation experiments are conducted to assess the value of the ship arrivals predicted by RF model on the improvement on daily operation planning of berth allocation and quay crane assignment in the GYCT terminal. And the results show that incorporating the predicted ship arrivals by RF model is beneficial to improve the performance of operation planning of GYCT terminal.