A data-driven approach for evaluating wind turbine degradation

This study proposes a data-driven methodology for assessing wind turbine degradation using SCADA data. A Random Forest regression model is trained to estimate an Aging Indicator (AgI) directly from operational variables such as normalized wind speed, pitch angle, and generated power, eliminating the need to reconstruct annual power curves for all turbines or rely on additional assumptions, such as Rayleigh distributions. To isolate long-term degradation patterns, the AgI time series is further analyzed using Seasonal-Trend decomposition via Loess (STL), which reveals a characteristic efficiency trajectory for this wind farm: an initial phase of performance improvement, followed by a plateau and then a gradual decline. In turbines operated under Wind Sector Management (WSM), sector-based operating restrictions can attenuate apparent degradation signals and extend plateau phases. Validation against traditional AEP-based assessments demonstrates good agreement, with evaluation metrics achieving R 2 of 0.994 ± 0.006 and an RMSE of 0.0010 ± 0.0006. The estimated degradation rates are also consistent with values reported in the literature, reinforcing the reliability of the proposed approach for long-term performance monitoring. • A Random Forest model estimates the Aging Indicator (AgI) from operational data. • Trained on one turbine and consistent across the entire fleet. • Less sensitive to outliers than AEP methods, stable with minimal filtering. • Wind sector management may extend plateau and reduce apparent degradation. • Validated against AEP and tested on another site, confirming generalization.