Data-driven stochastic nonlinear model predictive control for 6 DOF underwater vehicle under unknown disturbance and uncertainty

Abstract

Optimizing the operational performance of a submarine requires considering both control robustness and efficiency. Controlling an underactuated submarine in maritime environments is difficult due to the following reasons: 1) Unknown disturbances that increase the probability of constraints on timely operations. 2) Parametric uncertainties resulting from limitations in the surrogate hydrodynamic model and inaccurate hydrodynamic coefficients. This study presents a data-driven stochastic nonlinear model predictive controller that incorporates a 6 degree of freedom (DOF) hydrodynamic model of a submarine with an online learning-based model that describes unknown nonlinear additive components. The online-learning-based model was derived through sparse Gaussian process regression using fully independent training conditions. Since the submarine operates at moderate speeds, we introduce a non-parametric distributed model for each planar direction. An ellipsoidal terminal constraint is employed to stabilize the system. To demonstrate the improvements in the control tracking performance, we conducted a 3-dimensional path-following simulations using a generic submarine hydrodynamic model in the presence of unanticipated disturbances and modeling uncertainties, thereby demonstrating the superiority of the proposed controller. Additionally, the increase in uncertainties and errors observed during the path-following simulations was examined by considering the hydrodynamic maneuvering characteristics of the submarine.