MS029 - Optimal Control and Decision Making under Uncertainty from Digital Twins
Keywords: digital twins, dynamical systems, neural networks, optimal experimental design, partial differential equations, uncertainty quantification
The mini-symposium aims at addressing the “virtual-to-physical” leg of a digital twin framework, in which an updated model is used as a basis for optimal control of the physical system. As in the data assimilation task, quantifying uncertainties is crucial for safe control and decision-making. This manifests as forward models that are characterized by random parameters, leading to optimal control problems that take the form of stochastic optimization problems which are governed by (usually PDE) models with random parameters.
Specific challenges in this framework then include: (1) the development of effective algorithms for PDE-constrained stochastic optimization problems when considering high dimensional random parameter spaces (as in discretizations of infinite dimensional fields) and/or complex models; (2) the introduction of measures that quantify risks, which typically lead to non-differentiable objective functions when discretized, complicating the use of efficient optimization methods; (3) the consideration of the sequential and real-time natures of control problems, constructing numerical methods based on filtering, model reduction, or adaptive techniques; (4) the control of sensors/observations themselves when the physical system evolves, to better learn about the state/parameters of the system.
The objectives of the mini-symposium is thus to discuss recent developments and emerging applications in this multi-disciplinary control field, and thus initiate exchanges between members of various scientific communities (computational mechanics, applied mathematics, computer science, control theory, etc.). We anticipate contributions on the following topics:
- Optimal control for large-scale nonlinear dynamical systems governed by ODEs, PDEs, or represented by black-box models;
- Uncertainty quantification for control, guaranteed control synthesis;
- Stability analysis over moving horizons;
- Real-time, robustness, safety, or portability constraints on the control process;
- Model reduction, offline/online, multi-fidelity, or adaptive strategies;
- IMC or MPC controllers;
- Correct-by-design command;
- Data-driven control (e.g. using NNs), self-learning control algorithms;
- Optimal experimental design (OED);
- Applications of control to challenging scientific, engineering, and technological problems in aerospace, transport, biomedicine, climate, energy, manufacturing, robotics…
Specific challenges in this framework then include: (1) the development of effective algorithms for PDE-constrained stochastic optimization problems when considering high dimensional random parameter spaces (as in discretizations of infinite dimensional fields) and/or complex models; (2) the introduction of measures that quantify risks, which typically lead to non-differentiable objective functions when discretized, complicating the use of efficient optimization methods; (3) the consideration of the sequential and real-time natures of control problems, constructing numerical methods based on filtering, model reduction, or adaptive techniques; (4) the control of sensors/observations themselves when the physical system evolves, to better learn about the state/parameters of the system.
The objectives of the mini-symposium is thus to discuss recent developments and emerging applications in this multi-disciplinary control field, and thus initiate exchanges between members of various scientific communities (computational mechanics, applied mathematics, computer science, control theory, etc.). We anticipate contributions on the following topics:
- Optimal control for large-scale nonlinear dynamical systems governed by ODEs, PDEs, or represented by black-box models;
- Uncertainty quantification for control, guaranteed control synthesis;
- Stability analysis over moving horizons;
- Real-time, robustness, safety, or portability constraints on the control process;
- Model reduction, offline/online, multi-fidelity, or adaptive strategies;
- IMC or MPC controllers;
- Correct-by-design command;
- Data-driven control (e.g. using NNs), self-learning control algorithms;
- Optimal experimental design (OED);
- Applications of control to challenging scientific, engineering, and technological problems in aerospace, transport, biomedicine, climate, energy, manufacturing, robotics…
