On Ship Track Prediction with LSTM and Attention Mechanism

A method of ship track prediction based on combination of attention mechanism and long short-term memory is developed to cope with the complexity of inland river traffic that brought about by navigational environment and heavy traffic density. Valuable information is extracted from AIS data mining with the method. A LSTM encode-decode prediction model is built for process ship historical AIS data according to the sequential character of AIS position data. The temporal and spatial attention mechanism is introduced into the basic LSTM encode-decode prediction model to simulate the motion of ships sailing on their own and the impact of encountering situations. The loss function and output mode are defined and the attention-LSTM model of recurrent neural network type is completed. The model is trained and verified with existing AIS data sets.