A Study on Wind Power Curve Modeling Using Adaptive DBSCAN

Accurate power curve modeling is essential for reliable grid integration, but operational data often contain outliers caused by curtailment, turbine faults, and sensor errors. These outliers distort the wind speed power relationship and reduce prediction– accuracy. To address this, we propose an Adaptive DBSCAN outlier removal method designed to handle non-uniform data density. The approach first applies Zonal DBSCAN, which divides the data into wind speed segments using 3-quantiles and optimizes clustering parameters within each segment to detect outliers more effectively. A follow-up IQR refinement removes remaining extreme points. Using data from two wind farms, the method showed strong robustness across different conditions. When applied to polynomial regression, SVR, and spline interpolation, the cleaned data consistently improved model accuracy. Incorporating weather forecast data further demonstrated its usefulness for power generation forecasting and electricity market bidding. The proposed method provides a simple and practical preprocessing tool that enhances both operational decisions and economic performance for wind farm operators.