Alexandre Hache thesis defense

Tuesday 02.04.2025
Time:
From 10:00 to 12:00

Address:

Nantes Campus - Amphithéâtre Blaise Pascal

 Alexandre Hache from the Department of Automation, Production and Computer Sciences (DAPI) and LS2N laboratory will present his research about :

"Robust control of nonlinear systems based on a physical-neural hybrid LPV model"

 

Thesis defense notice 

See more

Abstract : This thesis lies at the crossroad between learning theory and control theory, proposing a data-driven methodology for modeling and controlling nonlinear dynamical systems. Drawing from the absolute stability theory,
and from a general representation of neural state-space models, several stability theorems are presented. Facing the limitations of traditional optimization approaches under LMI constraints for neural networks, we develop a complete theoretical framework for neural network parameterization, compatible with gradient algorithms and classical automatic differentiation tools. With the help of feedback linearization theory, a single-step learning method of an approximately linearizing controller and a reference model with guaranteed stability properties is presented. The theoretical results are validated on academic examples of disturbance attenuation, paving the way for more systematic use of neural networks in controllers’ design. 

Organizer(s)

Thesis co-acreditation from IMT Atlantique with the doctoral school SPIN.

 

Keywords : Neural State-Space Models, Absolute stability, Feedback Linearization, SDP Constraints

Published on 28.01.2025
Next events