A novel ship trajectory prediction method based on E-Informer

As the shipping industry advances and maritime traffic grows, accurate vessel trajectory prediction has become essential for route planning, navigation safety, and logistics optimization. To enhance prediction accuracy and efficiency, this paper presents an improved prediction method for complex inland waterways, termed Enhanced Informer (E-Informer), building on the foundation of the traditional Informer model. The key innovation involves replacing the original Informer’s decoder with Bi-LSTM, enabling bidirectional trajectory feature extraction. This design effectively integrates historical and future navigation data, enhancing the model's capacity to capture the global context of vessel movements. Moreover, a Multi-Query Attention (MQA) mechanism is introduced to optimize encoder-decoder interactions. By sharing keys and values within the attention mechanism, this approach not only accelerates decoding but also boosts real-time responsiveness while reducing computational complexity. Experimental results demonstrate that the E-Informer model outperforms the traditional Informer in prediction accuracy and decoding efficiency, making it particularly suitable for real-time trajectory prediction, intelligent navigation, and maritime emergency alerts in complex maritime environments.