A spatial-temporal attention method for the prediction of multi ship time headways using AIS data

Abstract

Ship Time Headway (STH) is the time interval between two consecutive ships arriving in the same water area. It serves as a crucial indicator for visually measuring the probability of ship congestion and the frequency of passage in busy waterways. Accurately predicting the STH is crucial for effective maritime traffic management. In this paper, we propose a deep learning method aimed at simultaneously predicting the STH in multiple water areas (multi-STH). This method integrates the Variational Mode Decomposition (VMD) algorithm with the Spatial-Temporal Attention Graph Convolution Network (STAGCN) to deeply capture the complex spatial-temporal features between STHs of each water areas. STH sequences were obtained from Automatic Identification System (AIS) for each reach, ensuring that these sequences remained numerically continuous on the same timeline. The VMD algorithm was employed to decompose the sequences into multi-feature inputs for the STAGCN, training the model in conjunction with the inland waterway traffic network to capture the patterns of variation in STH between the water areas. Extensive experiments demonstrate that the proposed prediction method surpasses the accuracy and robustness of other existing methods, exhibiting excellent prediction performance in the STHs of various waterways. The multi-STH prediction study accounts for the inherent correlation between inland waterways, substantially improving prediction efficiency compared to single-waterway STH prediction. This study may have the potential to provide useful support for traffic management. This may be of practical significance in enhancing the safety of inland waterways navigation.