Load Estimation in Onshore Wind Farms Using Surrogate Modeling and Generic Turbine Models

Abstract. This article investigates the development and application of surrogate models, based on slightly adapted generic turbine models, for predicting loads on real-world wind turbines. A small set of aeroelastic simulations provided training data for both Polynomial Chaos expansion and Gaussian Process regression models, which were trained to predict blade loads, tower accelerations, and their respective seed-to-seed variability. To evaluate the practical suitability of these models a case study was performed. Here, the surrogate models were applied to predict blade loads and tower accelerations respectively, using five years of SCADA data from an onshore wind farm. While the models approximated the real-world turbine behavior with a reasonable accuracy, the prediction quality varied across the different turbines in the park and was further influenced by factors such as the turbine's operational years and diurnal patterns suggesting a correlation with the turbulence intensity. Despite some limitations, the findings support the practicality of developing surrogate models for enabling efficient load estimations.