Prediction of offshore wind turbine wake and output power using large eddy simulation and convolutional neural network

Abstract

Predicting offshore wind turbine wake and output power is crucial for optimizing wind farm layouts and maximizing wind energy production. In recent years, several Computational Fluid Dynamics methods have been developed to predict wind turbine wake and output power and demonstrated good performance compared with traditional analytical models. However, Computational Fluid Dynamics often involve high computational costs in offshore wind farm design because a wide range of offshore wind conditions need to be considered for turbines with different inter-turbine spacings. To ensure both the fidelity and efficiency for predicting offshore wind turbine wake and output power, Large Eddy Simulation and Convolutional Neural Network are utilized in this study. The Large Eddy Simulation effectively integrates the Actuator Line Method and Discretizing and Synthesizing Random Flow Generation to generate wake velocity, wake turbulence intensity, and output power for a stand-alone turbine under different incoming wind speeds and turbulence intensities. Using the generated dataset, Convolutional Neural Network effectively captures the relationship between inputs and outputs for the stand-alone turbine. The predicted wake data for the turbine can then act as input to estimate the output power density and wake characteristics of a downstream turbine. This process can be iteratively applied to predict the wake and output power of each subsequent turbine in a wind farm, supporting the identification of optimal inter-turbine spacing. The proposed method is illustrated using a utility-scale 5 MW wind turbine. The results show that the errors of predicted output power for a stand-alone wind turbine and multiple wind turbines are blew 3 %.