A Marine Traffic Pattern Prediction System using Graph Attention Network and Bi-directional Long Short-Term Memory Network

In recent days, Marine activities are rapidly increasing in various sectors such as migration, logistics etc. and it is required to developa robust system for assisting the helmsman/driver in navigation. Therefore, it is essential to track the location of ships to avoid collision and in assisting the ships which diverts to different ship lanes. Hence, predicting location of ships is crucial to encounter dangerous and risky situations by developing ship trajectory prediction model. This paper illustrates a novel approach to predict ships’ trajectory patterns by combining of different Deep Learning (DL) techniques. Initially, the cohort is collected from Automatic Identification System (AIS) data and then firstly, Principal Component Analysis (PCA) is utilized for filtration and then scaling and interpolation is incorporated. Further, for trajectory pattern recognition Graph convolutional networks (GCNs) model is employed and finally a hybrid model using two networks Graph Attention Networks (GAT) with Bi-directional Long Short Term Memory Networks (Bi-LSTM) namely GAT-Bi-LSTM is proposed for the prediction of different marine traffic patterns. Therefore, the performance of the proposed model is determined in terms of Mean Absolute Error (MAE), Root Mean Square Error(RMSE), Average Displacement Error (ADE), and Final Displacement Error (FDE).