Improvement Interpolation Method for Vessel Trajectory Prediction based on AIS Data

Abstract

An acceptable shipping monitoring system should be able to provide early detection of vessel accidents. One of the keys to early detection is trajectory prediction. Even though the prediction results depend heavily on the historical data of the vessel trajectory, there are often missing values in the trajectory data due to disturbances in the Automatic Identification System (AIS) data transmission process. Therefore, this study proposes a preprocessing method that includes data cleaning, trajectory extraction, and a combination of Linear interpolation methods for straight-shaped trajectories and Cubic Spline interpolation for curved-shaped trajectories. Test results involving three different trajectories showed that the Gated Recurrent Units (GRU) method produces a smaller Root Mean Square Error (RMSE) value on linear interpolation. Visually, however, the GRU method with linear interpolation has a deficiency in the curved-shaped trajectories. Our studies involving the Bidirectional GRU (BiGRU), Long Short Term Memory (LSTM), and Bidirectional LSTM (BiLSTM) methods showed that the model built using the proposed interpolation method has a lower RMSE value. This study emphasizes that good predictions of vessel trajectories based on AIS data requires an additional process, namely the interpolation of vessel trajectories according to their shape.