Posters

Abstract

Accurate turbulence intensity characterization is essential for IEC 61400-1 site classification and turbine selection in repowering projects. However, many wind farms lack standard deviation measurements in SCADA systems, recording only maximum, minimum, and mean wind speeds over 10-minute intervals. This work presents two validated approaches for IEC turbulence classification using limited SCADA data, evaluated across 250+ measurement sites spanning onshore, offshore, and complex terrain conditions. A baseline Gaussian model assumes wind speed within 10-minute periods follows a Normal distribution, relating extrema to standard deviation through an optimized scaling parameter. This approach achieves 80.1% IEC classification accuracy with near-zero bias. A machine learning model implemented using Histogram-Based Gradient Boosting combines instantaneous 10-minute measurements with statistical features extracted from the 1-hour temporal window surrounding the analyzed 10-minute period. After correction of a systematic bias, this model achieves 82.6% IEC classification accuracy with robust generalization across turbulence classes (A+, A, B, C). Both models demonstrate absolute errors below 7% in 90% of evaluated cases. The framework enables cost-effective turbulence assessment using existing SCADA infrastructure, avoiding expensive and time-consuming measurement campaigns. Model selection guidelines are provided based on deployment context and accuracy requirements. Keywords: Turbulence intensity, SCADA data analysis, IEC 61400-1 classification, machine learning, wind farm repowering, site assessment