Model-free adaptive predictive control with unscented Kalman filter for hovercraft speed and heading stabilization

Abstract

This paper develops a novel unscented Kalman filter-model free adaptive predictive control (UKF-MFAPC) method for an underactuated hovercraft speed and heading stabilization under the influence of measurement noise. First, the proposed method is based on the compact form dynamic linearization (CFLD) technique, which involves an analysis of the tight format dynamic linearization method. This approach specifically addresses the unique non-real-time scalability of the heading control subsystem, rendering the CFDL method unsuitable. Through redefining the linear combination of lateral velocity and heading as output to fulfill all assumptions of CFDL methods, this scheme effectively mitigates the lateral velocity vibration problem of the hovercraft. Secondly, this controller introduces the measurement noise during the data collection process into the optimization function and combines the unscented Kalman filtering method with data-driven models, the influence of error suppression noise on the controller is suppressed, ensuring that the output can still be controlled within a certain range of expected values. This paper also delves into the convergence of tracking errors and examines the stability of bounded inputs and outputs. Finally, the noise suppression effect and stability of the UKF-MFAPC scheme are demonstrated through a comparative analysis of simulation results with proposed model-free methods.