In view of the challenges of the energy transition, and in particular the upcoming change from synchronous
machines to grid-forming inverters, methodological research on the collective dynamics of power grids is
highly topical Existing simulation environments are often not state of the art in terms of numerical methods
and the use of high-performance computing (HPC) and GPUs This prevents their use in large sampling
studies, which are necessary for the use of machine learning (ML), for example Furthermore,
methodological and mathematical advances are often difficult or impossible to implement in existing
software, which means they cannot be validated on realistic models and ultimately do not make it into
practice For methodological research, the flexibility of the implementation of models, as well as the
performance of simulations, is crucial The use of A1 and ML requires large amounts of simulation data as
input In order to even enter research on hybrid Al methods that directly combine physical simulations and
ML, it is necessary to implement the models in programming languages conceptualized for this purpose
Existing software tools also reach performance limits in practice Due to the limited performance of today's
simulation software, not all potentially relevant contingencies can be considered in time-critical situations
This project aims to close this gap by developing a software suite designed to quickly and effectively
integrate methodological innovations while being able to simulate realistic dynamic models of the power
grid PIK's tasks include coordination of the project, community building and workshops, training materials
and tutorials, development of backend and frontend, and initial research as model studies.
