A modified model predictive control based on B-spline fitting

Abstract

A reference signal is the target for a plant to track. When the reference signal is incomplete over a prediction horizon for model predictive control, a constant prediction of a reference is generally used to take place of the unknown reference signal. However, the complement of the constant reference prediction would lead to a discontinuity if the reference were not a constant signal. Moreover, the plant output signal is not supposed to follow a discontinuous reference especially for a tracking problem. In this paper, a model predictive control method based on B-spline fitting is presented. The B-spline fitting is used to interpolate the known reference signal and then a B-spline extension or extrapolation is employed to extend the reference curve in a continuous and smooth way. The new B-spline-treated reference then takes part in the optimization process of the model predictive control to generate the optimal input signal. The smooth extension could be closer to the actual trend of the reference, so it improves the performance of the model predictive controller. Simulation results show that this method works well when the prediction horizon is not large.