Ing Ind - Inf (Mag.)(ord. 270) - MI (473) AUTOMATION AND CONTROL ENGINEERING - INGEGNERIA DELL'AUTOMAZIONE
052354 - DATA DRIVEN CONTROL SYSTEM DESIGN
093061 - ADAPTIVE SYSTEMS AND LEARNING
In control and filter design problems, one deals with systems that often are affected by uncertainty so that their behavior is not deterministically known and, possibly, it is also time varying. In this situation, data-driven control design schemes can be adopted, where design is not based on a fixed model of the system, but rather modeling is progressively updated based on data obtained from the actually observed system response and the better control action to achieve the desired goal is learned through experience. This course aims at illustrating some of the main techniques in data-driven control and at demonstrating their effectiveness in counteracting the presence of uncertainty; correspondingly, advance identification and learning techniques to acquire knowledge on the system dynamics from measured data will be introduced and discussed.
Risultati di apprendimento attesi
Lectures and exercise sessions will allow the student to:
- recognize the situations where the design of control systems can profitably take advantage of learning from data and from the observation of the system behavior;
- learn how the wealth of information contained in the data can be processed and exploited for control and system design purposes;
- understand the assumptions and functioning principles underlying some common data-driven control design techniques;
- decide which schemes are more profitable depending on the application at hand, and predict the impact of algorithmic choices and data-preprocessing as for the robustness and effectiveness of the designed control scheme.
Uncertainty, data, adaptation and robustness
Uncertainty, robustness, and adaptation in control. Principles of data-driven control. On-line and off-line schemes. Direct and indirect schemes.
Indirect data-driven control schemes
The self-tuning paradigm. Certainty equivalence principle. Identification resumed: least-squares identification, recursive identification, forgetting factor. Main properties of the self-tuning scheme: convergence of the model estimate, stability and optimality. Examples.
Direct data-driven control schemes
Principles of direct data-driven control. Taxonomy of existing approaches (IFT, VRFT,…). The VRFT (Virtual Reference Feedback Tuning) approach: principles. Frequency interpretation of PEM identification methods. Optimality and sub-optimality of the VRFT controller, optimal data pre-filtering. VRFT in presence of additive disturbances. Instrumental Variable identification. The VRFT toolbox of MATLAB.
Advance topics on data-driven control systems design
Data-driven optimization: the scenario approach. Bayesian optimization and control. Optimal control and reinforcement learning
Basic knowledge in system theory and control and in probability (this knowledge is offered by fundamental courses offered in the programme of typical first level engineering degree). Though not strictly required, basic knowledge on learning and system identification (as offered by the course the Model identification and data analysis) is recommended.
Modalità di valutazione
The final exam is a written exam, constituted by four questions. The duration is 2h.
The first three questions are theoretical open questions on topics which have dealt with in the whole course programme. These questions are intended to indicate the level of knowledge and comprehension that has been reached by the student. Possibly, one of these question can be a numerical exercise, which is suited to evaluate the obtained capability of applying the obtained knowledge for the resolution of some problems.
In the fourth question the student is required to address a situation, which, though related to the topics of the course, has not been dealt with directly in the classes. This question is aimed at highlighting the student's ability to develop links within the various topics of the course and to apply the obtained knowledge in new innovative contexts.
Notes and papers provided by the lecturer Note:
Additional information available on the course web-site http://corsi.dei.polimi.it/IMAD/SAA/
S. Bittanti, Identificazione dei modelli e sistemi adattativi, Editore: Pitagora Editrice
S. Bittanti, Teoria della predizione e del filtraggio, Editore: Pitagora Editrice
S. Bittanti (ed.), Simulazione, identificazione, controllo: il caso di uno scambiatore di calore, Editore: Pitagora Editrice
Nessun software richiesto
Tipo Forma Didattica
Ore di attività svolte in aula
Ore di studio autonome
Laboratorio Di Progetto
Informazioni in lingua inglese a supporto dell'internazionalizzazione
Insegnamento erogato in lingua
Disponibilità di materiale didattico/slides in lingua inglese
Disponibilità di libri di testo/bibliografia in lingua inglese
Possibilità di sostenere l'esame in lingua inglese
Disponibilità di supporto didattico in lingua inglese