Control-oriented wind turbine load estimation based on local linear neuro-fuzzy models

Abstract

Abstract To enhance energy production, the wind energy industry builds increasingly larger wind turbines (WT) in terms of hub height and rotor diameter. This enlargement results in higher structural loads, which emphasizes the importance of load-reducing WT control alongside the nominal control of power and rotational speed. Generally, it is not possible to directly measure these structural loads, so a controller needs to include their online estimation. Here we extend a baseline WT state estimator in the form of an Extended Kalman Filter (EKF) to include two unaccounted loads. These loads are a change of thrust in the main bearing and a yaw moment in the rotor hub. We incorporate these loads via data-driven local linear neuro-fuzzy models (LLNFM). These LLNFMs can represent nonlinear relationships while maintaining limited complexity. We use alaska/Wind to generate the underlying regression data and to perform the state estimation both in simulation. The regression data consists of nominal WT operation under turbulent conditions for different average wind speeds. We further superimpose the individual pitch angles to excite the WT. The evaluation is performed in a different scenario, using another turbulent wind field and a realistic noisy sensor configuration. While we can estimate the change of thrust in the one per revolution (1P) frequency range, we achieve more precise results for the yaw moment due to the incorporation of sensors with a strong correlation to it. The estimation of the change of thrust and the yaw moment appear to be sufficiently precise for load-reducing WT control in practically relevant frequency ranges.