Parameter Estimation of Autonomous Underwater Vehicles Dynamics. A Data-Driven Optimization Framework

Accurate parameter estimation in ship dynamics models is pivotal for enhancing navigation precision, optimizing control systems, and ensuring maritime safety. This study explores two distinct methodologies: the “Grey Box” approach, where the governing equations of ship dynamics are known but certain parameters remain unidentified, and the “Black Box” approach, which relies entirely on trajectory data without any prior knowledge of the underlying mathematical models. Leveraging machine learning techniques, specifically Long Short-Term Memory (LSTM) networks, alongside optimization algorithms like L-BFGS-B, this research aims to evaluate and compare the efficacy of both approaches in estimating model parameters and predicting ship trajectories. The findings demonstrate that while the Grey Box approach benefits from incorporating physical laws for parameter tuning, the Black Box method offers flexibility in modeling complex, nonlinear dynamics purely based on empirical data. The integrated use of these methodologies provides a robust framework for enhancing the accuracy and reliability of ship dynamics models, contributing significantly to maritime engineering applications.