Enabling More Resilient Grids With Physics and Data-Driven ML Virtual Sensors
, Probabilistic Modeling Specialist, Siemens Energy
Siemens Energy is developing physics-machine learning surrogate models that are accurate and fast enough to integrate into operation control systems to enable more resilient power grids. Operating a power grid is a complicated task that requires complex controls. For many substation assets (like a transformer), it's not possible to quickly predict thermal behaviors with adequate accuracy during a short “overloading” situation. But, as simulations show, this should be perfectly acceptable if the duration and frequency of these events are properly managed to prevent critical hotspot temperatures within the assets. If such “controlled overloading” were allowed and the behavior evaluated with the help of virtual sensors in real time, then many grid congestion problems could be resolved with a software upgrade instead of costly infrastructure upgrades. We'll discuss several models built using the PhysicsNeMo SDK and share technical aspects of our process.