Citation: Zhang, Z.; Santoni, C.;
Herges, T.; Sotiropoulos, F.;
Khosronejad, A. Time-Averaged
Wind Turbine Wake Flow Field
Prediction Using Autoencoder
Convolutional Neural Networks.
Energies 2022, 15, 41. https://
doi.org/10.3390/en15010041
Academic Editors:
Luis Hernández-Callejo,
Sergio Nesmachnow and
Sara Gallardo Saavedra
Received: 15 November 2021
Accepted: 17 December 2021
Published: 22 December 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Time-Averaged Wind Turbine Wake Flow Field Prediction Using
Autoencoder Convolutional Neural Networks
Zexia Zhang
1
, Christian Santoni
1
, Thomas Herges
2
, Fotis Sotiropoulos
3
and Ali Khosronejad
1,
*
1
Department of Civil Engineering, Stony Brook University, Stony Brook, NY 11794, USA;
zexia.zhang@stonybrook.edu (Z.Z.); christian.santoni@stonybrook.edu (C.S.)
2
Wind Energy Technologies, Sandia National Laboratories, Albuquerque, NM 87185, USA; therges@sandia.gov
3
Mechanical & Nuclear Engineering Department, Virginia Commonwealth University,
Richmond, VA 23284, USA; sotiropoulosf@vcu.edu
* Correspondence: ali.khosronejad@stonybrook.edu
Abstract:
A convolutional neural network (CNN) autoencoder model has been developed to generate
3D realizations of time-averaged velocity in the wake of the wind turbines at the Sandia National
Laboratories Scaled Wind Farm Technology (SWiFT) facility. Large-eddy simulations (LES) of the
SWiFT site are conducted using an actuator surface model to simulate the turbine structures to
produce training and validation datasets of the CNN. The simulations are validated using the
SpinnerLidar measurements of turbine wakes at the SWiFT site and the instantaneous and time-
averaged velocity fields from the training LES are used to train the CNN. The trained CNN is then
applied to predict 3D realizations of time-averaged velocity in the wake of the SWiFT turbines under
flow conditions different than those for which the CNN was trained. LES results for the validation
cases are used to evaluate the performance of the CNN predictions. Comparing the validation LES
results and CNN predictions, we show that the developed CNN autoencoder model holds great
potential for predicting time-averaged flow fields and the power production of wind turbines while
being several orders of magnitude computationally more efficient than LES.
Keywords:
convolutional neural network; wind turbine; wake flow predictions; large-eddy simulation
1. Introduction
In a wind farm, turbine wake interactions cause power losses and may increase fatigue
loads on downwind wind turbines [
1
,
2
]. Therefore, the accurate prediction of turbine wakes
is an important consideration in wind farm layout optimization, which can improve the
efficiency of power production and reduce the overall levelized cost of energy. As a result,
extensive efforts have been made on analytical and numerical models for the estimation of
turbines wake [3–6].
Due to the simplicity and low computational cost, engineering models are widely
used to predict wake flows and optimize wind farm power production, especially in
industrial applications. The very first and extensively studied model was proposed by
Jensen [
7
]. This model was derived from mass conservation, assuming a top-hat shape
distribution of velocity deficit in the wake. However, the top-hat wake shape assumption is
an oversimplification of the actual wake flow, which can be represented more accurately by
a Gaussian distribution [
8
–
10
]. Furthermore, more complex real-life characteristics of wake
flows have also been considered to improve the accuracy and flexibility of the Gaussian
models, including the double-Gaussian type velocity profile of the near wake [
11
,
12
], three-
dimensional effects [
13
,
14
], more accurate models for turbulence intensities [
15
], wind
turbine yaw offset [
16
], atmospheric stability, and Coriolis force [
17
]. Although these
models are efficient, the accuracy varies significantly from case to case [
18
,
19
], especially
in the near wake region [
12
,
20
]. In addition, wake overlapping effects are not accurately
described, as shown by Archer et al. [21].
Energies 2022, 15, 41. https://doi.org/10.3390/en15010041 https://www.mdpi.com/journal/energies
