DT-DeviationState-Open: an open telemetry dataset for the deviation-detection module of a digital twin of a distributed system with energy facilities

A reproducible 10-minute dataset for the deviation-detection module of a digital twin of a distributed system with energy facilities, built from the openly licensed Penmanshiel wind-farm SCADA archive (Cubico Sustainable Investments, Zenodo 10.5281/zenodo.5946808, CC-BY 4.0). The distributed system is a wind farm of five turbines (network nodes); the task is binary classification of abnormal/forced turbine states from telemetry. Labels are taken from the independent SCADA status logs (IEC operating categories), so they are not derived from the features. Two aligned versions: base_v1 (5 raw telemetry features, 95,407 rows) and improved_v2 (11 features adding the digital-twin expected-state residuals: expected_power, power_residual, power_ratio). A classifier family (KNN, Logistic Regression, Decision Tree, GBM, MLP) is trained on each; the twin residuals lift 4 of 5 models, the best (GBM/MLP) reaching F1 about 0.89 at ROC-AUC about 0.98 (MLP F1 0.806 to 0.894). The improved_v2 twin reference is fitted on training rows only, so the held-out metrics carry no train-to-test leakage; the label comes only from independent SCADA status logs. Supplements the author's digital-twin thesis (ITMM-2025, 10.34185/1991-7848.itmm.2025.01.080) and the anomaly-detection article (2024).