Physics-Informed AUV Modeling Method and Application Under Current Disturbances

In this paper, an improved method for modeling the dynamics of an Autonomous Underwater Vehicle (AUV) is proposed, incorporating the impact of ocean currents on AUV navigation. Initially, the concept of Physics-informed Neural Networks (PINN) is introduced, which integrate the AUV dynamics equations with deep neural networks to establish an AUV model capable of long-term forecasting under calm water conditions. By integrating the actual navigation data of the AUV with the established dynamic model in calm water, the current estimation is derived, which is then utilized to adjust the output values of the calm water model. Through the implementation of the Model Predictive Control (MPC) algorithm based on the modified AUV dynamics model, optimal control strategies for trajectory tracking under the influence of current disturbances are developed. The simulation results demonstrate that the proposed modeling approach can accurately forecast the AUV's navigation status in current conditions. Furthermore, the MPC controller, based on this approach, can proficiently and consistently accomplish AUV trajectory tracking tasks in ocean current environments. This research provides a novel perspective on AUV dynamic modeling and control, with potential implications for various related research fields.