Abstract: To address the issue of low accuracy in ship trajectory prediction in complex navigable waters,this paper proposes a GRU-Attention-BiLSTM model for ship trajectory prediction. In the encoder part,the Gated Recurrent Unit（ GRU） is employed to capture temporal features in trajectory sequences. The decoder adopts a Bidirectional Long ShortTerm Memory Network（ Bi LSTM） integrated with an Attention mechanism to adjust the weights of data features. The model input is based on the longitude,latitude,speed and heading of the ship at the historical moment,and the ship density in the water area after median filtering smoothing is introduced as an additional feature. Using Automatic Identification System（ AIS） data from the core port area of Ningbo-Zhoushan Port in March 2024,the model was trained and validated.Quantitative and qualitative comparisons with GRU,LSTM,Seq2 Seq-LSTM,Attention-BiLSTM,and Transformer models demonstrate that the proposed model achieves superior prediction performance across different prediction durations and navigation scenarios.