Research on the classification of ship encounter scenarios based on CAE-LSTM
Abstract: To tackle the challenge of recognizing similar ship encounter scenarios under multi-ship interference coupling and dynamic evolution, this paper proposes a classification method that combines a Convolutional Auto-Encoder (CAE) and a Long Short-Term Memory (LSTM) recurrent neural network model. First, a method for extracting ship encounter scenarios considering spatiotemporal proximity constraints is designed, enabling the extraction of numerous real ship encounter scenarios from historical AIS data for subsequent classification. Then, by setting a time window and rasterizing the scenarios, a CAE-based model is constructed to characterize spatial interference of ships in the scenarios. Further, an LSTM network is used to learn temporal evolution features, achieving a low-dimensional spatiotemporal vector representation of ship encounter scenarios. Finally, hierarchical clustering is applied to classify different ship encounter scenarios based on these low-dimensional spatiotemporal vectors. The proposed method is validated through extensive experiments using data from Ningbo-Zhoushan Port, and the results show that this method can effectively extract real ship encounter scenarios and accurately identify similar scenarios. This research provides robust support for a deep understanding of ship encounter scenarios and the mining of similar ship behavior patterns.
Submission Number: 56
Loading