MKG-GNN:Maritime Knowledge Graph and GNN Framework for ship speed forecasting in port

Accurate ship speed prediction is critical for optimizing operational efficiency and ensuring navigational safety within complex port environments. The current speed prediction mainly relies on machine learning and deep learning methods. However, these methods generally have semantic gaps and failing to capture the semantic correlation of the port environment. We propose a Graph Neural Network (GNN) ship speed prediction framework based on Maritime Knowledge Graph (MKG), which is implemented in four steps. First, the framework initially utilizes an innovative Bi-directional Long Short-Term Memory (Bi-LSTM) model, which can be a unique two-way mechanism for the missing Automatic Identific Ation System(AIS) data interpolation. Secondly, the knowledge graph modeling of various entity relationships is carried out in combination with the core relationships of definition. Then use the knowledge graph embedding method RotatE to process the combination relationship in the knowledge graph after modeling. Finally, the framework integrates the embedded knowledge graph with the improved GNN model to achieve accurate port ship speed prediction. Extensive evaluations demonstrate that our framework significantly outperforms state-of-the-art baselines, reducing the Mean Absolute Error (MAE) by approximately 47.3% compared to DLinear and NLinear under the three typical scenarios of complex maneuvering, mooring and departure. Compared with the advanced PatchTST, this method has achieved a performance improvement of about 55.6% in the MAE index. Compared with the traditional timing prediction LSTM, our method has improved by about 83.1% in MAE. The ablation experiment further confirms that the removal of the knowledge graph module will lead to a significant decline in performance, which confirms its key role in capturing complex spatial-temporal dependencies.