Ship trajectory prediction (STP) is essential for navigation safety and maritime traffic management. A prediction model is developed that fuses ship motion features with hydrometeorological data through an enhanced bidirectional long short-term memory (Bi-LSTM) network. Trajectories are first grouped into traffic patterns via DBSCAN clustering and origin–destination analysis. Representative trajectory centerlines are extracted using Gaussian process regression (GPR) to characterize motion. Hydrometeorological variables along each trajectory are then computed by trilinear interpolation, yielding an integrated feature set that combines movement and hydrometeorological features. On this basis, an enhanced Bidirectional Long Short-Term Memory network (MS-Bi-LSTM) with a new MS compound loss is proposed. Validation is conducted using AIS and hydrometeorological data from the Qiongzhou Strait, China. Superior positional accuracy and interval-prediction performance over benchmark models are demonstrated. The approach provides theoretical insight and practical utility for improving ship safety in complex waters and advancing intelligent maritime surveillance. • Fuses motion and environmental data using an MS-Bi-LSTM network for prediction. • Identifies patterns via DBSCAN-OD clustering and extracts centerlines with GPR. • Employs a novel MS compound loss function. • Demonstrates superior accuracy in the Qiongzhou Strait.
Liu et al. (Tue,) studied this question.