Target Ship Trajectory Prediction Based on WOA-BIGRU

To further improve the accuracy of Maritime Autonomous Surface Ships (MASS) in predicting target ship trajectories, a ship trajectory prediction algorithm based on Bidirectional Gated Recurrent Unit (BIGRU) neural network is proposed. Historical data from the Automatic Identification System (AIS) is collected to build a trajectory database. The model is optimized using the Whale Optimization Algorithm (WOA) to minimize the mean square error between the expected output of the BIGRU network and the actual output, aiming to find the optimal network hyperparameters for precise prediction of the target ship trajectory. Additionally, the Dropout network structure is introduced to describe the prediction uncertainty of the model. By randomly deactivating neurons in the WOA+BIGRU core network, multiple sets of similar prediction results are obtained using the same trajectory sequence as input. The uncertainty of ship trajectory prediction is quantified based on their statistical mean and variance. Experimental results using AIS data from Shanghai Port show a model fitting goodness of 99.89%, and comparison with other neural network models demonstrates higher prediction accuracy of the proposed model.