کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Novel Data-Driven Koopman-Based MPC with Virtual Inputs for USV Trajectory Tracking under Unknown Dynamics and Unmodeled Environmental Disturbances

نویسندگان :

Ashkan Jelodar¹ Khalil Alipour² Bahram Tarvirdizadeh³ Ahmad Kalhor⁴ Mohammad Ghamari⁵ Majid Sorouri⁶

1- دانشگاه تهران 2- دانشگاه تهران 3- دانشگاه تهران 4- دانشگاه تهران 5- Department of Electrical Eng, California Polytechnic State University 6- Department of Electronic Engineering, Maynooth University, Maynooth, Co. Kildare, Ireland

کلمات کلیدی :

Koopman Operator

چکیده :

This article introduces a data-driven Koopman-based model predictive control (MPC) framework for trajectory tracking of unmanned surface vehicles (USVs) under dynamics that are unknown to the controller and in the presence of unmodeled environmental disturbances (e.g., waves, currents, and wind). Using Koopman operator theory, the proposed approach learns a finite-dimensional lifted linear predictor from data, enabling MPC design without requiring an explicit physics-based model in the controller. A critical obstacle in conventional Extended Dynamic Mode Decomposition (EDMD) for Koopman linearization is the orientation-dependent directional coupling of surge (u) and sway (v) inputs, which induces numerical instabilities, poor convergence, and inaccurate approximations, particularly under disturbances. To surmount this, we innovate a virtual input transformation that projects body-frame velocities onto fixed global axes using a yaw-dependent rotation matrix, thereby decoupling rotational effects, enhancing data quality in unknown environments, and facilitating the discovery of a low-dimensional lifted linear model with good prediction accuracy (e.g., prediction error below 9% on validation data; see Section IV for the metric definition). This linear model enables a computationally efficient convex MPC optimization with control rate penalties that promote smooth actuator operation and robust tracking under disturbances. Simulation results demonstrate accurate trajectory tracking and robust behavior under disturbances, while retaining the computational advantages of convex optimization.