A Time Series Forecast Method for Vessel Trajectory Prediction

Abstract

In recent years, alongside the progress of marine vessel information technology, the scale of vessel-related data has grown exponentially. At the same time, maritime monitoring based on vessel data has achieved unprecedented development, so how to effectively manage and supervise the marine operations of vessels is now a widely concerned issue. By predicting the future trajectories of vessels, vessel behavior can be assessed to avoid potential hazards. This paper establishes a pre-processing model of vessel data based on the time series data of vessels. According to the characteristics of vessel data, cleaning, noise reduction, and trajectory extraction are conducted to the data followed by interpolation. Trajectory similarity evaluation is conducted with a time series similarity measurement method, and a trajectory clustering model based on DBSCAN is constructed from the trajectory similarity information. Later, this paper proposes a time series prediction model based on Attention and LSTM. The prediction model adopts an Encoder-Decoder structure. The model takes the time series characteristics of a vessel trajectory as the input to predict the future trajectory of a vessel.