Machine-learning dataset for physically consistent early stability assessment of a grid-following and grid-forming PV–BESS plant

Machine-learning dataset derived from 476 electromagnetic-transient simulations of a 100 MWp PV and 60 MWh / 35 MW BESS plant with grid-following or grid-forming (VSM) BESS control, seven grid-strength levels (SCR 0.44–3.52) and 34 disturbance cases (PV drop, load rejection, islanding, temporary and permanent three-phase faults). Files:- features.csv: base features and ordinal labels (0 stable, 1 alert, 2 unstable) for every window length, noise level and noise realization- physics_features.csv: fourteen physical stability indicators The descriptions of the features and the risk directions of the indicators (feature_list.txt and physics_indicators.txt) and all code are available at https://github.com/[username]/pcsa-pv-bess-stability. The learning set contains 466 runs (132 stable, 213 alert, 121 unstable) after excluding ten runs with pre-event oscillation. The raw EMT waveforms are available from the corresponding author on reasonable request.