Budapest University of Technology and Economics, Faculty of Electrical Engineering and Informatics

    Belépés
    címtáras azonosítással
    vissza a tantárgylistához   nyomtatható verzió    

    Artifical Intelligence Based Control

    A tantárgy neve magyarul / Name of the subject in Hungarian: Mesterséges intelligencia alapú irányítások

    Last updated: 2024. február 27.

    Budapest University of Technology and Economics
    Faculty of Electrical Engineering and Informatics

    Electrical engineering major
    MSc degree program
    Control and Vision     Systems main specialization block 
    Course ID Semester Assessment Credit Tantárgyfélév
    VIIIMB06   2/1/0/v 5  
    3. Course coordinator and department Dr. Harmati István,
    Web page of the course https://edu.vik.bme.hu/
    4. Instructors Dr. habil. Harmati István
    5. Required knowledge Mathematics (Linear algebra, analysis, gradient-based numerical optimization), Control technology
    6. Pre-requisites
    Kötelező:
    NEM
    (TárgyEredmény( "BMEVIIIMA09", "jegy" , _ ) >= 2
    VAGY
    TárgyEredmény("BMEVIIIMA09", "FELVETEL", AktualisFelev()) > 0)

    A fenti forma a Neptun sajátja, ezen technikai okokból nem változtattunk.

    A kötelező előtanulmányi rend az adott szak honlapján és képzési programjában található.

    7. Objectives, learning outcomes and obtained knowledge
    The aim of the course is for students to gain knowledge about the latest methods of controlling and identifying complex systems using artificial intelligence methods, which are also used in practice. Students will learn about the concept and theoretical background of the following most common artificial intelligence methods:
    • Fuzzy systems
    • Genetic algorithms
    • Neural networks
    • Neuro-fuzzy systems
    • Swarm intelligence methods
    • Reinforcement learning
    The subject shows how the above-mentioned methods can be used primarily (but not exclusively) to solve control engineering, system modeling and optimization problems using modern computer science-supporting programming platforms (mainly MATLAB).

    8. Synopsis
    Detailed topics of the lectures:
     
    1. Basics of fuzzy systems. The concept of the fuzzy system, the theoretical foundations of fuzzy inference. The structure, rule base and algorithm of regulations operating on the fuzzy principle. MacVicar-Whelan meta-rules. (2 weeks)
    2. Construction of genetic algorithms. Genetic operators: selection, recombination, mutation, back substitution, migration. Controller design with genetic algorithm. (1 week)
    3. Linear and non-linear parameter estimation. Batch and recursive parameter estimation procedures for linear and non-linear system models. (1 week)
    4. Clustering procedures. Theoretical foundations of grid partitioning, subtractive clustering, fuzzy c-mean clustering, structure of algorithms (1 week).
    5. Construction of feedforward shallow neural networks, learning by error backpropagation. Basics of deep learning methods. Autoencoders, stochastic neural networks, convolutional networks in control tasks. Feedback (RNN, LSTM) networks for solving dynamic tasks (2 weeks)
    6. Identification with adaptive Neuro-fuzzy systems, structure of the method, tuning rules, ANFIS. (1 week)
    7. Adaptive fuzzy control. Nominal and supervisory control design, indirect (based on a model) and direct (not using a model) adaptive control, stability testing and parameter tuning rules. (2 weeks)
    8. Basics of reinforcement learning. Prediction and control of known and unknown/large Markov decision processes: Dynamic programming, Monte Carlo, Temporal Difference based learning, Sarsa, Q-learning. Basics of deep reinforcement learning: DQN, REINFORCE, Actor-Critic networks in prediction and control. (2 weeks)
    9. Swarm intelligence methods. Construction of ant colony algorithms and their applications for solving discrete optimization problems. The theoretical background of particle swarm optimization and the steps of the algorithm. Optimization based on swarm intelligence methods, system identification and control design. (1 week)
     
    The detailed topics of the exercises:
     
    1. Fuzzy controller design in Matlab-Simulink environment using Fuzzy Toolbox. (1 week)
    2. Determination of PID controller parameters using genetic algorithm in Matlab-Simulink environment (1 week)
    3. System identification with linear and non-linear parameter estimation in the Matlab environment (1 week)
    4. Implementation of clustering and ANFIS methods in Matlab environment. (1 week)
    5. Prediction of failure of railway safety equipment using Matlab Deep Learning Toolbox. (1 week)
    6. Control of a nonlinear system with reinforcement learning in the Matlab environment using the Reinforcement Learning Toolbox. (2 weeks)
    ​

    9. Method of instruction The course material is presented in lectures and exercises. Lectures and exercises alternate at the pace of the material. In the exercises, the theoretical material presented in the lectures is deepened in the form of calculation examples and case studies.
    10. Assessment
    In study period: The knowledge of the course material is measured once during the stuy period with a written midterm in closed form. The condition for obtaining the signature is to pass the midterm at least at a sufficient level.
    In exam period: Obtaining the signature is a condition for admission to the exam. The exam consists of a written performance evaluation and the  the result achieved in the midterm. There is no way to improve the result of midterm during the exam period.
    The grade obtained for the subject is determined 20% from the result (score) achieved in the midterm and 80% from the exam.
    11. Recaps There is one retake in the study period or on the retake week.
    12. Consultations During the study period, it is primarily during the consultation hours of the subject's instructors, or at a pre-arranged time if needed. During the exam period, after electronic consultation, on the working day before the exam. Instructors reserve the right not to respond to student letters/messages where the required information is clear based on the subject's data sheet or website.
    13. References, textbooks and resources

    B. Lantos: Fuzzy systems and genetic algorithms, 2002, Műegyetemi kiadó

    M. Dorigo: Ant Colony Optimization, MIT University Press Ltd., 2004

    R. S. Sutton, A. G. Barto: Reinforcement Learning: An introduction, MIT Press, 2018

    P. Kim: MATLAB Deep Learning, Apress, 2017

    14. Required learning hours and assignment
    Kontakt óra42
    Félévközi készülés órákra36
    Felkészülés zárthelyire20
    Házi feladat elkészítése-
    Kijelölt írásos tananyag elsajátítása16
    Vizsgafelkészülés36
    Összesen150
    15. Syllabus prepared by István Harmati, Dr. habil.  Associate Professor, Dept. of Control Engineering and Information technology