Model predictive control with on-line optimal linearisation

This paper describes a nonlinear Model Predictive Control (MPC) algorithm with on-line optimal linearisation. Unlike the classical MPC algorithms with successive model linearisation at the current operating point of the process (using the Taylor´s series expansion), the best possible linear approximation of the nonlinear predicted output trajectory is repeatedly found in the discussed approach. The algorithm is computationally efficient, because it requires solving linear equations and quadratic optimisation problems. It is demonstrated that for a benchmark system the described MPC algorithm gives significantly better control quality than the classical MPC approach with model linearisation.