Wake state estimation of downwind turbines using recurrent neural networks for inverse dynamics modelling

Abstract

Presented in this work is a novel approach to estimate absolute lateral wake center position on the rotor plane of a waked turbine using turbine load and operating state information. The approach formulates the estimation of the absolute lateral wake position as an inverse dynamics problem and utilizes a recurrent neural network to model the inverse mapping between the wake center position and select turbine output channels. The technique is validated on experimental data collected from experiments at the Scaled Wind Farm Technology (SWiFT) facility and numerical simulations of the site in the wind farm simulator FAST.Farm. Estimator performance and analysis of optimal conditions for estimation are discussed.