Evaluation and comparison of MPC algorithms applied to simulated processes

Abstract

Model predictive controller is one of the most advanced control approaches which because of its good features like: ability to explicitly include constraints in its formulation, multivariable control, ability to look in to the future and act proactively has become popular in the industry. Although it was first introduced in the control of power production and in the petroleum industry, it has recently among others have been used in the automotive industry and medical applications like the artificial pancreas. It is thus worth making some research on the modern control system with good prospects for the future. This thesis focuses on evaluating and comparing MPC algorithms (linear and nonlinear model predictive controller algorithms) applied to simulated processes. An air-heater model and a model for an anaerobic digestion reactor are considered as case studies. Linear and nonlinear model predictive controller algorithms are developed and used to control each of the models. Robustness, stability and computation time of each of the controller algorithms under disturbance, measurement noise and model errors are evaluated and compared.