Model predictive control (MPC) with integral action; Reducing the control horizon and model free MPC

Abstract

Model Predictive Control (MPC) is the most widely used strategy in process industries due to remarkable features. It has the capability to control the non-minimum phase, unstable processes and handle the constraints in a systematic way. MPC with integral action is an effective method to achieve the offset free control which can remove the unknown slowly varying process and measurement noise respectively. In this thesis, a multivariable four-tank process has been developed for simulation experiments and it is controlled at two operating conditions i.e. minimum and non-minimum phase setting. The mathematical models are constructed from the both physical and simulation data. Theoretical background of the state space model based MPC is described and the deviation variables are used to achieve the integral action in MPC. The proposed optimal controller has been implemented to control the level in lower tanks. The ‘quadprog’ function and ‘if-else’ technique are demonstrated to handle process constraints in MPC with integral action. The execution time for simulation is reduced using ‘if-else’ method compared to ‘quadprog’ function. The states are estimated by using the Kalman filter. A comparison in reducing control horizon in optimal control is also performed. The decentralized PI controller has been implemented to control the four-tank process and results are compared with MPC method. Deterministic and Stochastic system identification and Realization ‘DSR’ algorithm has been proposed to formulate model free MPC. A linearized state space model is identified by the ‘DSR’ method and used in MPC algorithm. The proposed optimal control is more robust and faster than the traditional PI controller. Simulations are performed in MATLAB software.