Léo Simpson

Tuesday, January 18, 2022, 11:00 - 11:59

"Online"

In this talk, I will present my project for my Ph.D. about numerical methods for learning-based control of thermal networks. The application I am interested in consists of controlling the temperature of an unknown system through a fluid whose temperature is controlled by Temperature Control Units (TCUs). The overall system is composed of two parts: the units, in which a temperature model is available; and a thermal network in which there is no available model of the dynamics.

A supervisory control method will be discussed and illustrated by some first results. This method consists of separating the task into two distinct problems: the first one is a (possibly nonlinear) control problem with linear constraints corresponding to the control of the TCUs. The second is a control problem under uncertainty corresponding to the unknown thermal network. For the first one, a classical MPC method is sufficient. However, for the second, we need a data-driven method tailored to the knowledge we have about the structure of the system

In this context, I will also present a characterization of thermal networks using positive or compartmental systems. Furthermore, I will briefly present the problems of identification and stabilization of systems with such properties.