AIS Data-Driven Maritime Traffic Flow Prediction and Density Visualization Using Multitime Scale Temporal Feature Fusion Network

Abstract

Maritime traffic flow prediction is essential for the development of intelligent transportation systems in the maritime domain. The widespread deployment of automatic identification system (AIS) sensors generates vast amounts of real-time trajectory data, which are crucial for traffic state perception and vessel position tracking. This study aims to improve maritime traffic state prediction by leveraging AIS data. The existing methods often face two key challenges: insufficient consideration of traffic flow characteristics across different time scales, which limits the comprehensive capture of temporal features, and difficulties in estimating the uncertainty in ship position distributions. To address these challenges, we propose a multitime scale temporal feature fusion network (MSTFFN) model. This model enhances the transformer architecture for maritime traffic flow prediction by extracting multiscale temporal features that encapsulate the dynamic nature of traffic patterns. Additionally, a Gaussian distribution process is employed to effectively visualize traffic density. Experiments on real-world datasets demonstrate the superior performance of the MSTFFN model in traffic flow prediction tasks. Compared to baseline models, such as transformer, GRU, long short-term memory (LSTM), bidirectional LSTM (Bi-LSTM), and CovLSTM—each improved through multitime scale feature fusion—the proposed MSTFFN achieves superior predictive accuracy. Moreover, the advanced visualization of traffic density facilitates more intuitive and efficient maritime traffic management, thereby enhancing the application of AIS data across ship-to-shore and management operations in the maritime industry.