Real-time optimization of wind farms using modifier adaptation and machine learning

Abstract

Coordinated wind farm control takes the interaction between turbines into account and improves the performance of the overall wind farm. Accurate surrogate models are the key to model-based wind farm control. In this article a modifier adaptation approach is proposed to improve surrogate models. The approach exploits plant measurements to estimate and correct the mismatch between the surrogate model and the actual plant. Gaussian process regression, which is a probabilistic nonparametric modeling technique, is used in the identification of the plant–model mismatch. The efficacy of the approach is illustrated in several numerical case studies. Moreover, challenges in applying the approach to a real wind farm with a truly dynamic environment are discussed.