Lecture
Model Predictive Control Talks
by Prof. Eric C. Kerrigan (Imperial College London) and Assoc. Prof. Martin Klauco (Slovak University of Technology)
Location: Technical University of Denmark, Building 101, Room S09
Date and time: Monday September 16, 2024. 10:00-11:30
Everybody is welcome.
10:00-10:45
Integrate your residuals while solving dynamic optimization problems
Professor Eric C. Kerrigan, Imperial College London, London, UK
Abstract:
Many optimal control, estimation and design problems can be formulated as so-called dynamic optimization problems, which are optimization problems with differential equations and other constraints. State-of-the-art methods based on collocation, which enforce the differential equations at only a finite set of points, can struggle to solve certain dynamic optimization problems, such as those with high-index differential algebraic equations, consistently overdetermined constraints or problems with singular arcs. We show how numerical methods based on integrating the differential equation residuals can be used to solve dynamic optimization problems where collocation methods fail.
10:45-11:30
Neural Network-Based Approximation of Model Predictive Control: Enabling Lightweight Deployment and Cryptographic Integration
Associate Professor Martin Klauco, Slovak University of Technology, Bratislava, Slovakia
Abstract:
We present a machine learning-based toolchain that approximates Model Predictive Control (MPC) by deriving an explicit control law represented as an interconnected neural network. This approach offers several key advantages for MPC applications. First, the explicit representation enables deployment on platforms with limited computational power and memory, such as PLC devices, etc. Second, this neural network-based control law provides a possibility to be integrated with cryptographic methods, opening new possibilities for secure and nearly-optimal control in sensitive environments, for example preventing future Stuxnet incident.