Operator Learning for Power Systems Simulation

TL;DR

Time-domain power-system simulations become intractable with high renewable penetration due to ultra-fast dynamics requiring very small time steps $\Delta t$. The authors propose operator-learning surrogates that are discretization-invariant and benchmark three architectures—DeepONets, Fourier Neural Operators, and Latent Neural ODEs—to map initial disturbed trajectories to future system evolution. They evaluate zero-shot super-resolution and regime-generalization on a SMIB system, revealing that time-step-invariance enables generalization to finer steps, with LNODE variants showing robust performance under coarse training and adaptive step control, and mixed training improving robustness across stable and unstable regimes. The work provides a scalable, data-driven tool for rapid stability assessment in renewable-rich grids, aiding climate-change mitigation efforts.

Abstract

Time domain simulation, i.e., modeling the system's evolution over time, is a crucial tool for studying and enhancing power system stability and dynamic performance. However, these simulations become computationally intractable for renewable-penetrated grids, due to the small simulation time step required to capture renewable energy resources' ultra-fast dynamic phenomena in the range of 1-50 microseconds. This creates a critical need for solutions that are both fast and scalable, posing a major barrier for the stable integration of renewable energy resources and thus climate change mitigation. This paper explores operator learning, a family of machine learning methods that learn mappings between functions, as a surrogate model for these costly simulations. The paper investigates, for the first time, the fundamental concept of simulation time step-invariance, which enables models trained on coarse time steps to generalize to fine-resolution dynamics. Three operator learning methods are benchmarked on a simple test system that, while not incorporating practical complexities of renewable-penetrated grids, serves as a first proof-of-concept to demonstrate the viability of time step-invariance. Models are evaluated on (i) zero-shot super-resolution, where training is performed on a coarse simulation time step and inference is performed at super-resolution, and (ii) generalization between stable and unstable dynamic regimes. This work addresses a key challenge in the integration of renewable energy for the mitigation of climate change by benchmarking operator learning methods to model physical systems.

Figures (2)

• Figure 1: Zero-Shot Super-Resolution: This figure illustrates the true time series of the system angle ($\delta$) and frequency ($\omega$) dynamics in solid lines and the operators' predicted trajectories in dotted lines. The plots start with the segment of the true trajectory in a solid blue line used as the input.
• Figure 2: Generalization across different dynamical regimes: Training data has $0\%$ (blue) and $20\%$ (orange) unstable trajectories. The green dotted line is where $P_{m1} = P_m$, i.e., no disturbance, and the red dotted line is the critical point where trajectories become unstable. The mean absolute scaled error is calculated using a baseline-prediction of the previous time step over 20 runs.