Nonlinear model predictive control

This thesis addresses the control of general nonlinear processes using a nonlinear model predictive controller (NLMPC). Nonlinear capabilities are lent to the controller in the form of nonlinear process and disturbance models. The nonlinear models that are used in this thesis are of two types: first principle models that are based on the physical laws; and empirical models that are based on the input-output plant data.; The first part of the thesis gives a general overview of the nonlinear model predictive control algorithm and presents various techniques available for the identification of nonlinear systems. The structure of the input-output models used in this work is of polynomial Auto Regressive with eXogenous inputs (ARX). The second part of the thesis focuses on the analysis of the input-output models from the control point of view in MPC framework. Two industrial case studies are presented where these input-output models and the nonlinear control algorithm are used to control highly nonlinear processes. In MPC a quadratic objective function is minimized to arrive at the control inputs. The global optimality of the solution is sometimes difficult to ascertain if there are nonlinearities present in the objective function and constraints. An algorithm is proposed which guarantees the global solution to the MPC objective function using polynomial ARX models. The third part of the thesis focuses on developing a comprehensive controller package which is capable of tackling industrial control problems with various features including nonlinear state and parameter estimation.