Adaptive Generalized Predictive Control for Ship Autopilot

Ship autopilot is always characterized as designs for none-linear and uncertain system (variable hull shape, draft and speed of vessel, et al.), and exacerbated by environment, i.e. the effects of wind, sea-state, and tide, which is the facts challenging the classic PM controller. In this paper, a new adaptive autopilot based on extended recursive least (ERL) and generalize predictive control (GPC) is described. The identification of vessel model parameters is always contaminated with noise. ERL adopted in this paper is based SVD(Singular Value Decomposition), which obviously not only improve the convergence rate, numerical stability of ERL, but also can handle color noise without any transformation and provide much more precise identification results. Accurate estimation of vessel parameters from noise is completed with SVD-based ERL. Propulsion losses suffer from unexpected motion of rudder arose by the wave disturbance. The paper proposes that GPC with observer polynomial suppress the disturbance and minimize induced motions. The autopilot performs well in variable working environment which is verified in simulation and semi-physical platform experiment.