VO 328.036/UE 325.025 Adaptive and Predictive Control
Motivation and Overview
This course is focused on the design and analysis of model-predictive control (MPC). Even though classic PID control is widely used in industry due to its simplicity, many complex industrial processes could be run significantly more efficiently by employing advanced control methods.
Model predictive control (MPC) is one of these methods that has become a standard in the process industries in chemical plants and oil refineries in recent years. Model predictive control is based on a dynamic model of the process, which is used to predict its future states and outputs. In this sense MPC has the ability to anticipate future events and to take appropriate control actions accordingly. The principle of MPC is that it determines the current control signal by optimisation while keeping future control moves in account. This is achieved by optimizing over a finite time-horizon, but only implementing the current timeslot (receding horizon control). A remarkable and unique feature of MPC is the possibility to directly consider constraints in the determination of the control signal, so that the process can be controlled optimally while staying within safety limits. Such constraints might consider internal process states (e.g. maximum motor voltage, maximum temperatures, etc.) as well as limits on the control signal itself (e.g. actuator saturation, max. actuator speed).
While initially MPC was limited to relatively slow processes (e.g. in petrochemical plants) due to limited computer performance, today’s vastly increased computation speed and its cheap availability render MPC applicable in virtually all control applications. The course covers this exciting field of control theory and shows typical applications.
This course starts by giving a basic overview on selftuning control, gain-scheduled control, and adaptive control methods and typical use cases. After that, the course’s main part is devoted to the theory of modelpredictive control. The main ideas of predictive control are explained, and various MPC formulations and the related optimization problems are developed in detail, so that students are empowered to develop or modify MPC algorithms on their own. To this avail, relevant aspects and properties of the underlying constrained optimization problems and methods to solve them (e.g. active-set methods) are discussed. Finally, stability theory of model predictive control is addressed.
The course treats single-input-single-output (SISO) systems as well as multiple-input-multiple-output systems, enabling students to gain understanding of the nature of sophisticated real-world process control problems. A broad set of turnkey-ready MPC design examples and analysis algorithms are provided for the students in MATLAB/Simulink to illustrate the covered topics and to enable the students to study and solve typical industrial control tasks. Control problems and results provided by industry partners highlight the control performance attainable via MPC.