Multi-factor influence-based ship trajectory prediction analysis via deep learning

Abstract

The trajectory prediction research based on deep learning methods shows more substantial competitiveness than classical ones in the context of big data analysis methods. However, the relevant literature fails to explain the collective impact of multiple influential factors identified from Automatic Identification System (AIS) data, including latitude, longitude, Course Over Ground (COG), and Speed Over Ground (SOG). To fill in this research gap, six classical deep learning methods are newly employed to conduct ship trajectory prediction, taking into account multiple influential factors for the first time. Two real AIS datasets collected from water areas of high representation are chosen to test and analyse the performance of the six deep learning models against seven indexes. The experimental results reveal that both the traditional factors of longitude and latitude and the newly incorporated ones of SOG and COG play a key role in trajectory prediction. Moreover, the effect of SOG on the accuracy of prediction results is greater than that of COG. Furthermore, the advantages and disadvantages of the six trajectory prediction models revealed by the experimental results provide useful insights into the best-fit method under different circumstances of traffic management involving Maritime Autonomous Surface Ships (MASS).