Wind power modelling using physical models and gaussian processes

Abstract

Accurate modeling of power production in wind energy systems is essential for optimizing real-time operation and meeting technical or economic objectives. This work compares two approaches: a parametric one, based on analytical expressions of the power coefficient CP(λ,β), and a non-parametric one, which uses Gaussian Processes (GP) to probabilistically model the relationship between operating variables and generated power. Parametric models are efficient, interpretable and useful when the system dynamics are known, performing well under different scenarios. In contrast, GP models offer greater flexibility, integrate uncertainty and adapt to complex patterns in the data. We analyse the fit of both approaches with real wind turbine data and evaluate their performance using metrics such as Root Mean Square Error (RMSE) and computational cost. The results show that both methods are complementary for modelling the behaviour of wind turbines in different contexts.