A Multi-Step Nonlinear Predictive Controller - An Introduction

A new model predictive formulation to control MIMO nonlinear processes is presented. It directly extends the model predictive concepts that have been exploited for the control of linear systems (e.g., IMC and Q/DMC). This new algorithm uses a multi-step predictor, unlike the algorithms by Economou, et al. [2] and Li and Biegler [5], which were limited to a single predictive step. This paper summarizes the features of the multi-step predictor which exhibits superior performance over single-step methods. Furthermore, this multi-step algorithm does not require iterative calculations, as required by Economou, et al., and is more efficient for each predictive step. The algorithm also easily handles constraints involving the state variables and manipulated inputs.