lunduniversity.lu.se

Denna sida på svenska This page in English

Seminars and Events at automatic control

All seminars are held at the Department of Automatic Control, in the seminar room M 3170-73 on the third floor in the M-building, unless stated otherwise.

 

Lic Defense: Fethi Bencherki: Adaptive Control of Positive Systems

Disputation

From: 2025-02-14 10:15 to 12:00
Place: Lecture hall MH:G (Gårdingsalen), LTH
Contact: anders [dot] rantzer [at] control [dot] lth [dot] se


Date & Time: February 14th, 10:15-12:00
Location: Lecture hall MH:G (Gårdingsalen), LTH
Zoom: 
Title: Adaptive Control of Positive Systems
Opponent: Postdoctoral Fellow Yassir Jedra, Massachusetts Institute of Technology
Examiner: Professor Bo Bernhardsson, Lunds universitet
Supervisor: Professor Anders Rantzer, Lunds universitet
 

Abstract: The analysis, control, and design of complex networks and systems represent an essential and ever-expanding field of study. At its foundation lie mathematical models that capture complex physical phenomena while remaining amenable to both analysis and design. These models prove instrumental in representing diverse systems such as power grids, cellular telecommunication networks, large-scale manufacturing processes, and computer systems. However, as technological and societal needs grow, these systems continue to increase in both complexity and scale, making their controlling tasks increasingly challenging. Traditional control approaches often fail to capture the uncertainties inherent in the modeling process, resulting in significant performance deterioration. Furthermore, these methods frequently suffer from poor scalability, rendering them unsuitable for large scale systems.

In response to these challenges, robustness and adaptation have emerged as critical design principles. Robustness ensures resilience against unmodeled dynamics, while adaptation addresses unexpected events and changes in the dynamics. Scalability, on the other hand, can be achieved by deriving localized control laws, thereby alleviating the limitations of centralized control approaches. The increasing affordability and accessibility of data measurements further provide a foundation for advancing these objectives. This has paved the way for data-driven approaches, which are now gaining traction as a preferred strategy.


The first paper of this thesis, Robust Simultaneous Stabilization via Minimax Adaptive Control, certifies the existence of adaptive minimax controllers. These controllers are a superior class of robust adaptive nonlinear controllers that can stabilize the plant in the presence of large model uncertainties.

The second paper, Robust adaptive data-driven control of
positive systems with application to learning
in SSP problems, presents another type of robust adaptive controllers. This controller is designed for a positive systems class, where the property of positivity is exploited to enable the scalability of these control solutions.

The third and final paper, Data-Driven Adaptive Dispatching Policies
for Processing Networks, applies the framework from the second paper to processing networks. In this case, the processing rates of the different units are initially unknown. The paper analyzes the online performance of the learning-based adaptive policy.