"SCADA_Physics-Guided Adaptive DBSCAN for Sensor Data Quality Assurance in Wind Turbine SCADA Systems"

"Raw wind turbine supervisory control and data acquisition (SCADA) records are contaminated by heterogeneous anomalies\u2014sensor faults, operational shutdowns, and grid\u0002mandated curtailment\u2014that severely degrade the wind turbine power curve (WTPC) used for performance assessment and condition monitoring. Existing screening methods rely on two\u0002dimensional wind-speed\u2013power analysis, which cannot discrim\u0002inate these physically distinct anomaly categories. This paper proposes a two-stage physics-guided framework. Stage 1 cross\u0002validates wind speed, pitch angle, and generator speed through a three-dimensional physics-based constraint matrix to flag implausi\u0002ble operating states. Stage 2 applies a Physics-Guided Adaptive DBSCAN (PG-DBSCAN) that incorporates the theoretical aerodynamicgradient and adaptively sets neighborhood parameters from local sensor noise characteristics, detecting subtle anomalies that evade conventional density-based methods. A parameter unification principle reduces effective tuning to a single interpretable threshold. PG-DBSCAN achieves a post-cleaning power curve R 2 = 0.994 (RMSE = 34.8 kW) and fidelity F = 0.742, outperforming five baselines; an ablation study attributes a 22.7% RMSE reduction to the gradient alignment criterion. Cross-site validation on an independent 2 MW turbine datasetconfirms the framework\u2019s generalizability"