Optimizing Maritime Vessel Trajectory Prediction Using Space-Based AIS Data and PSO-BiGRU

This paper introduces a sophisticated approach for predicting maritime vessel trajectories in the Seas around Papua Island, East Indonesia, leveraging space-based Automatic Identification System (AIS) data from LAPAN satellites. The study emphasizes enhanced navigation safety and efficiency within this strategically vital region and employs Particle Swarm Optimization-Bidirectional Gated Recurrent Unit (PSO-BiGRU) to refine the hyperparameters of neural network architectures. Additional models such as RNN, LSTM, GRU, and their bidirectional counterparts serve as comparative benchmarks. A pivotal innovation in this research is the use of PSO-based hyperparameter tuning, which strategically optimizes learning rates, neuron counts, optimizers, layers, and dropout. This is complemented by the integration of bidirectional layers that utilize both past and future data contexts, markedly boosting the accuracy of trajectory predictions. The PSO-BiGRU outshines other models, particularly on vessel MMSI 367344000, achieving a Mean Square Error (MSE) of 0.00003, an improvement of over 40% compared to the standard GRU, and up to 7 kilometers reduction in average Haversine distance, underscoring the model's substantial advancements over other setups. This research not only highlights the effectiveness of bidirectional layers in deep learning models but also showcases the profound capability of PSO in navigating the intricate parameter space of neural network architectures, providing significant enhancements for maritime operations and safety in the East Indonesian maritime domain.