Physics-Informed Neural Networks for Vessel Trajectory Prediction via Finite Difference Kinematic Modeling
This project proposes the development of a Physics-Informed Neural Network (PINN) framework for vessel trajectory prediction using Automatic Identification System (AIS) data collected from Canadian maritime regions. The approach integrates discretized kinematic motion equations—specifically Euler, Heun, and midpoint finite difference approximations—into the training of deep learning models to enforce physical consistency. By combining data-driven architectures (LSTM, GRU, CNN, TCN, Transformer) with physics-based loss functions, the project aims to improve the accuracy, stability, and interpretability of trajectory predictions, particularly in data-sparse or complex navigational contexts such as the Arctic and the Strait of Georgia. Three undergraduate interns from Brazil will contribute in the areas of physical modeling, deep learning integration, and data engineering. The research is hosted at Dalhousie University’s MAPS Lab, led by Dr. Gabriel Spadon, and forms part of an ongoing collaboration with the University of São Paulo. The project’s outcomes are expected to advance Canadian innovation in smart marine systems by enabling safer navigation, optimized logistics, and enhanced environmental protection.
View Full Project DescriptionGabriel Spadon
Universidade de São Paulo
Computer science
Education
Dalhousie University
Globalink Research Award