Field-SCADA measurement analysis of power-curve deviation and sector-dependent underperformance in an operating utility-scale wind turbine

Supervisory control and data acquisition (SCADA) data can support transparent power-curve screening, but residual calculation, sector aggregation and wind-speed-bin median fitting are established tools rather than algorithmic innovations. This study analyzes 50,530 ten-minute records from a public 2018 single-turbine dataset containing active power, wind speed, wind direction and a source-supplied theoretical reference. Negative active-power entries were clipped to zero and no synthetic records were generated. In a strictly later 30% chronological holdout, a standard 0.25 m s −1 bin-median baseline reduced MAE from 181.45 kW for the source reference to 137.16 kW (24.4%), with R 2 increasing from 0.909 to 0.933. This is a field-calibration benchmark, not a new predictive algorithm. Directional efficiency is heterogeneous in an omnibus label-permutation test ( p = 0.0005 ), but sector precision depends on exposure: the 90–120° sector has the highest point estimate (0.986; 95% moving-block-bootstrap confidence interval 0.835–1.091; n = 540 ), whereas the well-sampled 180–210° sector has 0.909 (0.886–0.927; n = 6975 ). The 8–12 m s −1 regime contributes 61.54% of the negative-deviation burden. Calendar-half sector ranks are correlated ( ρ = 0.888 ), supporting descriptive persistence but not causal attribution. The study provides a bounded reproducibility benchmark and identifies the density, status and multi-turbine data required for causal analysis.