Learning Model Predictive Control - WEUMLMP5

Goal(s)

The course explores modern techniques for optimisation, control, system identification and learning for control. It provides fundamental mathematical and computational tools for the analysis and design of dynamic systems.

Responsible(s)

Olivier SENAME

Content(s)

1. Convex Optimisation: Introduction to convex sets and functions, optimality conditions, duality and stochastic optimisation. Numerical optimisation methods. Use of solvers through Yalmip in MATLAB and Pyomo in Python.
2. Optimal Control and Predictive Control on a finite horizon (MPC): the concepts of optimal control, tracking problems and the construction of cost functionals on a finite horizon with a view to stability, admissibility, extension to non-linear systems, and tracking problems.
3. From Dynamic Programming to Reinforcement Learning and Learning MPC: Bellman iterations for finite and infinite horizon problems, Markov Decision Processes. Policy iteration and value iteration techniques. Connections between MPC and Reinforcement Learning for learning values and policies.

The course provides a theoretical and practical basis for the application of advanced tools in control, system modelling and machine learning for the dynamics and control of complex systems

Linear systems, linear algebra, basis on non linear systems

Test

Final exam (ET1): A written examination with a time limit (1 hour 30 minutes)
Continuous Assessment (CC1): Assessment of project work (continuous assessment) with a written report and/or oral defence

Calendar

The course exists in the following branches:

• Curriculum - Master inter MARS - Semester 9

Additional Information

Course ID : WEUMLMP5
Course language(s):

You can find this course among all other courses.