Geohash coding-powered deep learning network for vessel trajectory prediction using clustered AIS data in maritime Internet of Things industries

Abstract

The increasing traffic flow and frequent navigational activities in maritime transportation raise accidents, such as collision and grounding. To mitigate such safety issues, high precision and stable vessel trajectory prediction play an essential role in developing accident prevention techniques in maritime Internet of Things (IoT) industries. However, vessel trajectory prediction is a challenge because conventional deep learning technologies struggle to capture complex changes in vessel trajectories. To address this challenge, this study designs a novel prediction method (namely DBSCAN-GeoCLSTM) comprising two steps. The first step employs the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to perform clustering analysis on trajectories. The second step feds trajectory data from different categories into a Long Short-Term Memory (LSTM) optimization network with Geohash Coding (called GeoCLSTM netework) for training. This GeoCLSTM network encodes the trajectory points using Geohash Coding to construct one-hot vectors, which captures the positional correlation information among trajectory points. The GeoCLSTM model further introduces a dimensionality reduction module by converting the one-hot vectors into feature vectors. Subsequently, the GeoCLSTM model employs LSTM to use these feature vectors from multiple consecutive time slots to accurately predict trajectory data. Results of comparative experiments demonstrate that the proposed DBSCAN-GeoCLSTM method achieved accurate and stable predictions, and outperformed eight state-of-the-art methods in the vessel trajectory prediction task.