Discovering Power Grid Dynamics from Data Using Low-Rank Sparse Modeling

TL;DR

<3-5 sentence high-level summary> This paper tackles the challenge of estimating dynamic parameters in low-inertia power grids by developing Latent-SINDy (L-SINDy), which blends Singular Value Decomposition (SVD) with Sparse Identification of Nonlinear Dynamics (SINDy) to learn reduced-order models directly from time-series data. By projecting the high-dimensional grid state onto a small set of latent coordinates, the method performs sparse regression in the latent space to identify inertia and damping terms that govern system dynamics, and then reconstructs full-state behavior. The approach is validated on IEEE-118, IEEE-300, and a 2869-bus European grid, achieving accurate recovery of inertia constants and damping with significant computational savings and robust performance across scales. Open-source MATLAB code is provided for reproducibility, underscoring the method's potential for real-time monitoring and control in renewable-rich grids.

Abstract

The growing integration of renewable energy sources has significantly reduced grid inertia, making modern power systems more vulnerable to instabilities. Accurate estimation of dynamic parameters such as inertia constants and damping coefficients is critical, yet traditional model-based methods struggle with scalability and adaptiveness in large, low-inertia networks. This paper presents a novel data-driven framework that integrates Singular Value Decomposition (SVD) with Sparse Identification of Nonlinear Dynamics (SINDy) to estimate system parameters directly from time-series data. By reducing dimensionality before applying sparse regression, the proposed Latent-SINDy (L-SINDy) method mitigates overfitting while preserving essential system dynamics. The framework is validated on IEEE benchmark systems, including the 118-bus, 300-bus, and a large-scale 2869-bus European grid. The results demonstrate an accurate recovery of the inertia and damping values, with identified models that closely match the true dynamics of the system. The approach offers scalability, interpretability, and computational efficiency, highlighting its potential for real-time monitoring and control in renewable-rich grids.

Figures (15)

• Figure 1: Graphical illustration of LSINDy
• Figure 2: Singular Value decay plot of IEEE-118 bus system
• Figure 4: Original latent states and their SINDy predictions for the IEEE 118 Bus systems. The solid lines denote the actual latent states obtained by compressing the true data, and the dashed lines denote the SINDy predictions.
• Figure 5: Comparison of true and estimated values of avg. $\delta$ and avg. $\Delta\omega$ in IEEE-118 bus system.
• Figure 6: Error plots of the average $\delta$ and avg. $\omega$ error in IEEE-118 bus system.