IEEE 9-bus benchmark brings 20,000 transient stability scenarios to AI

A team led by Hussein Suprême, Martin de Montigny, and Arnaud Zinflou has published a new open benchmark dataset designed to accelerate machine-learning research in power system transient stability assessment. The dataset, now available on IEEE DataPort with a persistent DOI, contains 20,000 three-phase-to-ground fault scenarios simulated on the IEEE 9-bus system. Each scenario couples an AC power-flow operating point with detailed electromagnetic-transient simulations of the post-fault response. Crucially, every record provides the network as an attributed graph—nine buses, eighteen directed branches, with ten node features and twelve edge features—plus full rotor-angle and speed trajectories for all three generators, static machine constants, fault descriptions, and a binary center-of-inertia stability label.

This dataset directly addresses a key bottleneck in the field: the lack of open datasets that combine dynamic ground truth, network graph structure, and machine parameters. Wide load and generation scalings across eighteen fault locations produce a near-balanced distribution (48.96% stable, 51.04% unstable), making it suitable for training and benchmarking without heavy class-imbalance adjustments. All generation is deterministic through fixed seeds and public code, ensuring full reproducibility. Researchers can use it for stability classification, trajectory prediction, transient stability margin estimation, and critical-clearing-time calculations. It also enables fair comparison of topology-aware, physics-based, and hybrid learning methods. By bridging detailed simulation data with graph-structured ML inputs, this benchmark could help AI models become practical tools for real-time grid stability monitoring and early warning systems.