Abstract: With the continuous increase in seaborne trade volume, ship traffic density in port areas is rising, and navigation conditions in port waters are becoming more complex. Short-term ship traffic prediction in port waters is playing an increasingly critical role in ship traffic control and navigation safety management. To address the limitation of low accuracy in aggregate models, this paper, based on ship Automatic Identification System（AIS） data, employs a disaggregate method to construct a hybrid prediction model. This model combines the Long Short-Term Memory（LSTM） network with ships′ historical trajectories to calculate short-term ship trajectories in port waters. The counts of ships′ trajectories intersecting with an approach channel section are used to predict the short-term ship flow across the section. A numerical example from Ningbo-Zhoushan Port during June to December 2020 demonstrates that the forecasting accuracy of the proposed model reaches up to 80%, significantly higher than that of traditional aggregate models. The model developed here provides a technical foundation for ports to implement ship traffic control methods and improve channel utilization rates.