Friday, 15th of November
11h (room R2014, 660 building) (see location
Graph Neural Solvers for Power Systems
We propose a novel neural network architecture based on graph neural network that learns to simulate complex physical systems. Instead of imitating a « classical » solver, it learns by penalizing the violation of physical laws during training, which makes it a completely new and independent method to solve this problem. This novel approach enables a substantial gain in terms of computational time compared to classical methods. More specifically, we apply our method to a problem of power flow prediction on power grids.
All TAU seminars: here