<b>A Frequency Domain Approach to Predict Power System Transients</b>

The dynamics of power grids are governed by a large number nonlinear differential and algebraic equations (DAEs). To safely operate the system, operators need to check that states described these DAEs stay within prescribed limits after various potential faults. However, current numerical solvers often too slow for real-time system operations. In addition, detailed parameters not exactly known. Machine learning approaches have been proposed reduce computational efforts, but existing methods generally suffer from overfitting failures predict unstable behaviors. This paper proposes novel framework transients in frequency domain. intuition is although behavior complex time domain, there relatively few dominant modes Therefore, we learn constructing neural networks with Fourier transform filtering layers. System topology fault information encoded taking multi-dimensional transform, allowing us leverage fact trajectories sparse both spatial frequencies. We show approach does parameters, greatly speeds up prediction computations highly accurate different types.