Data‑Driven Emulation of Wind Turbine Pitch Control with an RBF Network Under Varying Turbulence Conditions

Abstract

Above rated wind speed, collective pitch control (CPC) is essential for wind turbines to regulate generator speed and power. This paper proposes a data-driven emulation framework that reproduces the input-output behaviour of the reference software OpenFAST CPC controller using a radial basis function (RBF) network with a sliding time-window feature extraction. Multi-scale statistical features—derived from generator speed error, power error, pitch angle, wind speed, and generator speed over a 30‑s window—are fed into an RBF network to predict the next-step collective pitch angle. The model is trained on OpenFAST simulation data under three load cases: DLC1.1, DLC1.3 and DLC1.6 datasets. Two configurations (Single‑DLC, Multi‑DLC) are compared. Closed‑loop validation under DLC1.1 and DLC1.3 datasets shows that the RBF emulator achieves high fidelity (R² > 0.99, RMSE < 0.5°) and closely tracks the baseline PI controller with negligible performance degradation. The proposed surrogate controller provides a lightweight, accurate alternative for scenarios where the original regulator is inaccessible or fast prototyping is required.