| Time: | May 28, 2024 |
|---|---|
| Download as iCal: |
|
Prof. Dinesh Krishnamoorthy
Department of Mechanical Engineering
Eindhoven University of Technology
Eindhoven, Netherlands
Tuesday 2024-05-28 4 p.m.
IST Seminar Room 2.255 - Pfaffenwaldring 9 - Campus Stuttgart-Vaihingen
Abstract
Model Predictive Control (MPC) problems are frequently cast and solved as parametric Nonlinear Programming (NLP) problems. NLP parametric sensitivities offers a computationally cheap and versatile framework for understanding how optimal solutions change with parametric variations. This talk delves into how parametric sensitivities can be leveraged to our advantage in two important classes of MPC paradigms, namely, learning-based MPC and distributed MPC. The first part of the talk presents a sensitivity-based data augmentation framework to efficiently generate several training data points that can used to learn the control policy or the value function that can be appended as cost-to-go functions. The second part of the talk explores how parametric sensitivities can be used to accelerate distributed MPC problems by exploiting the parametric nature of the subproblems from one iteration to the next.
Biographical Information
Dinesh Krishnamoorthy is an Assistant professor at the Department of Mechanical Engineering at TU Eindhoven, where he is a part of the Control Systems Technology group. Prior to this, he was a post-doctoral researcher at Harvard University. Dinesh received his PhD in Process Systems Engineering from the Norwegian University of Science and Technology (2019), MSc in Control Systems from Imperial College London (2012), and B.Eng in Mechatronics from the University of Nottingham (2011). Dinesh was also working as a Senior Researcher at Statoil Research centre between 2012-2016. Among others, Dinesh has received the Dimitirs. N. Chorafas Foundation Award, PhD Excellence Award from the European Federation of Chemical Engineers (EFCE), NTNU Faculty of Natural Sciences Best PhD Thesis Award, IFAC Young author award, as well as a Veni Early Career Talent Grant from the Dutch Research Council. His research interests include distributed optimization, optimal control, and data-driven optimization, with applications to energy systems.