Physics-Informed System Identification
May 30, 2023, American Control Conference 2023, San Diego
Registration: https://acc2023.a2c2.org/registration/
Organizers: Sivaranjani Seetharaman and Vijay Gupta, Purdue University
This workshop will bring together experts to discuss recent advances in physics-informed and control-relevant system identification of a wide spectrum of models including linear, nonlinear, Koopman-operator based, time-varying, networked, and neural ODE models, and their applications spanning robotics and energy networks, to epidemics and neuroscience.
System identification has long been a bedrock of modeling for control, with many attractive and efficient approaches being developed over the decades for identifying linear, nonlinear, time-varying, parameter-varying, switched, and PDE-based dynamical system models. In recent years, system identification has seen a lot of renewed excitement due to its relevance to learning and similar fields. Many control applications deal with physical systems that naturally display physical constraints such as conservation laws and symmetries that must necessarily be captured models obtained through system identification to provide physically meaningful results. At the same time, many dynamical systems also possess special physical properties such as passivity, dissipativity, monotonicity, or positivity, that can be exploited to facilitate elegant analysis and control designs.
In this context, this workshop will bring together experts to discuss recent advances in physics-informed and control-relevant system identification of a wide spectrum of models including linear, nonlinear, Koopman-operator based, time-varying, networked, and neural ODE models, and their applications spanning robotics and energy networks, to epidemics and neuroscience.
We have an exciting slate of experts on the topic. Topics to be covered include identifying a wide class of system properties in linear and nonlinear systems to facilitate data-driven control designs, identification of neural ODE models capturing another important physical property, namely Hamiltonian equations of motion, with wide-ranging applications in multi-agent control and robotics, learning of time-varying and nonlinear physics-constrained network models and parameters with applications to epidemics and neuroscience, and learning linear operator models of nonlinear systems capturing the property of dissipativity. Taken together, these talks will provide a broad overview of emerging frontiers in physics-informed system identification and various exciting application domains.
ACC 2023 Worksop 