Predicting Fault-Ride-Through Probability of Inverter-Dominated Power Grids using Machine Learning

TL;DR

This work addresses the challenge of evaluating dynamic stability in inverter-dominated power grids under many fault scenarios by framing stability as a probabilistic problem and training machine learning models to predict fault-ride-through probability $p_{frt}$. It introduces a synthetic dataset of 1,000 grids (70–80 buses) with grid-forming and grid-following components, and uses Sobol sampling and ambient forcing to generate post-clearance states and ride-through curves. Graph Neural Networks, particularly the topology-aware TAG variant, are shown to outperform non-graph models on synthetic data, and with regularization they generalize effectively to the IEEE 96-RTS test system, demonstrating potential for fast, large-scale probabilistic stability assessments in future grids. The study highlights the critical role of topology in stability and suggests a practical ML pathway for early planning and vulnerability identification in low-inertia, inverter-rich power systems.

Abstract

Due to the increasing share of renewables, the analysis of the dynamical behavior of power grids gains importance. Effective risk assessments necessitate the analysis of large number of fault scenarios. The computational costs inherent in dynamic simulations impose constraints on the number of configurations that can be analyzed. Machine Learning (ML) has proven to efficiently predict complex power grid properties. Hence, we analyze the potential of ML for predicting dynamic stability of future power grids with large shares of inverters. For this purpose, we generate a new dataset consisting of synthetic power grid models and perform dynamical simulations. As targets for the ML training, we calculate the fault-ride-through probability, which we define as the probability of staying within a ride-through curve after a fault at a bus has been cleared. Importantly, we demonstrate that ML models accurately predict the fault-ride-through probability of synthetic power grids. Finally, we also show that the ML models generalize to an IEEE-96 Test System, which emphasizes the potential of deploying ML methods to study probabilistic stability of power grids.

Figures (6)

• Figure 1: The structure of the power system generation algorithm shows the generation based on the chosen input parameters. Before the power grid is returned by the software, its behavior is validated to fulfill the stability criteria of real power grids.
• Figure 2: Training setup: The GNN gets the full power grid as input (bus features, line features and the topology) whereas the nodal features are the only input for the non-graph ML methods. In all cases, the outputs are the nodal fault-ride-through probability ($p_{frt}$).
• Figure 3: Example transients of the dynamical simulations, visualized by the frequency deviation from the reference frequency of 50 Hz and the voltage magnitude $V$. The plots visualize a stable configuration (top) and an unstable configuration (bottom), for which the voltage thresholds are exceeded. The dashed line in gray denotes the operational boundaries. The black transients mark the bus where the fault occurred. The time is plotted on a logarithmic scale to highlight the first time steps.
• Figure 4: The active and reactive power deviations $\Delta P$ and $\Delta Q$ resulting from the faults at three different buses of the IEEE test case, and the outcome of the corresponding simulations. Darker points indicate that the systems survives and returns to stable operation, whereas light points indicate a failure. Color distinguishes between the types of the shown buses, namely a load bus (top), a NF1 bus (mid), and a NF3 bus (bottom).
• Figure 5: The histograms of the $p_{frt}$ for the IEEE test case are on the left and for the 1 000 synthetic grids on the right. The rows depict the four different bus types. NF1-3 denotes the different normal form parameterizations. For the loads, a different scaling of the vertical axis is used.