MINI COURSE: Mouhacine Benosman

Senior Principal Research Scientist, Mitsubishi Electric Research Laboratories (MERL)

A hybrid approach to control: classical control theory meets machine learning theory

Until recently, one could classify control approaches into two main paradigms. The first is the classical machine learning (ML) control paradigm that heavily relies on data, e.g., classical approximate dynamic programming (ADP), reinforcement learning (RL) methods, deep neural networks (DNN), and deep RL. The main advantages of these methods are efficiency and flexibility due to the increased availability of data and computation power. On the other hand, these methods lack performance guarantees, such us stability, boundedness of signals, a.k.a., safety, and general robustness.

The second paradigm is the classical control theory approach, which relies on dynamical systems theory, e.g., robust control theory, adaptive control theory, Lyapunov-based design, etc. In this case the pros and cons are somewhat reversed. Some examples of advantages of this paradigm include the rigor of mathematical analysis and the performance guarantees in term of stability, boundedness of signals, and robustness. However, one of the main disadvantages of this approach is the lack of flexibility or generalizability since the model of the system must satisfy very specific properties.

In the past 10 or so years, several efforts have attempted to merge these two paradigms.  The result is what we refer to as Learning-based control methods. These methods use tools from classical control theory together with tools from ML theory. The aim is to design ‘hybrid’ learning-based controllers that take advantage of the flexibility of ML data-driven methods, while maintaining the stability, safety, and robustness guarantees from control theory.

This short course will concentrate on learning-based control methods. We first present recent results in the field of learning-based adaptive control, where classical model-based adaptive control methods are merged with data-driven estimation methods, e.g., extremum seeking control (ESC), Gaussian processes (GP) optimization, and reinforcement learning (RL). We then discuss some recent results on robust constrained model-based RL that use tools from nonlinear control theory to guarantee stability, robustness and safety. We will cover the main theoretical aspects of these approaches and finish the course with a few examples of industrial applications.

Short Biography :

Before coming to Mitsubishi Electric Research Laboratories (MERL) in 2010, Mouhacine worked at universities in Reims University, France and Strathclyde University, Scotland, and the National University of Singapore.

His research interests include modeling and control of flexible robotic manipulators, nonlinear robust and fault tolerant control, multi-agent control with applications to smart-grid and robotics, estimation and control of partial differential equations with applications to thermo-fluid models, learning-based adaptive control for nonlinear systems, and control-theory based optimization algorithms with application to machine learning.

Mouhacine has published more than 50 peer-reviewed journal articles and conference papers, and has more than 20 patents in the field of mechatronics systems control. He is a senior member of the IEEE, Associate Editor of the Journal of Optimization Theory and Applications, Associate Editor of the Journal of Advanced Control for Applications, Associate Editor of the IEEE Control Systems Letters, and Senior Editor of the International Journal of Adaptive Control and Signal Processing.



Professor of Engineering Science at the University of Oxford, UK

Data-driven battery health diagnosis in real-world applications

Accurate diagnostics and prognostics of battery health improves overall system performance. This allows industry to unlock value by detecting faults and improving maintenance and logistics, extending operational range, and understanding asset depreciation. However, battery aging is complex and caused by many interacting factors. Two key questions arise: first, how to handle modelling challenges, including parameter variability and nonlinearities, in methods for online estimation of state of health. Second, how to develop validated predictions of future health, where key issues include coping with variable usage scenarios, and cell-to-cell behavioural differences. This talk will discuss recent approaches to tackle some of these exciting topics, particularly focusing on diagnostics from field data, including the combining of non-parametric and parametric models to allow flexibility in model fitting from data, whilst retaining the benefits of equivalent circuit and physical models.

Short Biography:

David Howey is Professor of Engineering Science at the University of Oxford, UK. He leads a group researching on modelling and control of energy storage systems, with a particular focus on Li-ion batteries for electric vehicles and grid/off-grid storage. He received the MEng degree in Electrical and Information Sciences from the University of Cambridge in 2002 and his PhD from Imperial College London in 2010. Since 2010 he has co-authored 80+ peer-reviewed journal and conference articles, and 5 patents. He is an editorial board member of IEEE Transactions on Industrial Informatics and the new OUP journal Oxford Open Energy, and is co-founder of the Oxford Battery Modelling Symposium. He is the recipient of recent funding from EPSRC, InnovateUK, UKRI, Faraday Institution, Continental AG and Siemens, and he co-leads control and estimation tasks in the Faraday Institution Multiscale Modelling project. Howey is also academic lead for the £40m Energy Superhub Oxford that is building a transmission connected 50 MWh hybrid battery. He previously led the Faraday Institution “UK EV and Battery Production Potential” project (with McKinsey), and was academic lead in InnovateUK projects on battery re-use (EP/P510737/1) and solar home systems in Africa (EP/R035822/1), and a $1.2m Korean project on microgrids, plus Co-I in EPSRC projects TRENDS, FUTURE vehicles, STABLE-NET and RHYTHM. Professor Howey is co-founder of Brill Power Ltd., a company spun-out of his lab in 2016 focused on advanced battery management system topologies. They have raised significant early stage funding and adopted several patents from his group. Howey also won a Samsung GRO Award on modelling leading to two R&D contracts and a multi-year collaboration, with results patented by Samsung Electronics.


Professor of control and automation systems at the Universidad de los Andes, Colombia

The Role of Population Games and Evolutionary Dynamics in Control

Recently, there has been in the control community an increasing interest in studying large-scale distributed systems (LSDS). Several techniques have been developed, wishing to address the main challenges found in LSDS. One way to approach this type of problems is to use game-theoretical methods. Game theory shares some common points with control systems problems, in particular of distributed topology, where the interconnection of different elements (agents) leads to a global behavior depending on the local interaction of these agents. Evolutionary game theory (EGT) is one type of dynamic games that has been used to design distributed controllers for different applications like control of water systems, charging of electric vehicles, and synchronization of isolated microgrids. The aim of this talk is to present and discuss relevant advances and analytical methodologies in population games and evolutionary dynamics, and its applications for solving control problems.

Short Biography:

Nicanor Quijano (IEEE Senior Member) received his B.S. degree in Electronics Engineering from Pontificia Universidad Javeriana (PUJ), Bogotá, Colombia, in 1999. He received the M.S. and PhD degrees in Electrical and Computer Engineering from The Ohio State University, in 2002 and 2006, respectively. In 2007, he joined the Electrical and Electronics Engineering Department, Universidad de los Andes (UAndes), Bogotá, Colombia as an Assistant Professor. He is currently a Full Professor, the director of the research group in control and automation systems (GIAP, UAndes), and an associate editor for the IEEE Transactions on Control Systems Technology, the Journal of Modern Power Systems and Clean Energy, and Energy Systems. He has been a member of the Board of Governors of the IEEE Control Systems Society (CSS) for the 2014 period, and he was the chair of the IEEE CSS, Colombia for the 2011-2013 period. He has published more than 100 scientific papers (journal papers, international conference papers, book chapters), he has co-advised the best European PhD thesis in the control systems area in 2017, and he is the co-author of the best paper of the ISA Transactions, 2018. In 2021, he obtained the Experienced Research Award from the School of Engineering, UAndes. Currently his research interests include hierarchical and distributed network optimization methods for control using learning, bio-inspired, and game-theoretical techniques for dynamic resource allocation problems, especially those in energy, water, agriculture, and transportation.