Search for: All records

With the increased availability of experimental measurements aiming at probing wind resources and wind turbine operations, machine learning (ML) models are poised to advance our understanding of the physics underpinning the interaction between the atmospheric boundary layer and wind turbine arrays, the generated wakes and their interactions, and wind energy harvesting. However, the majority of the existing ML models for predicting wind turbine wakes merely recreate CFD-simulated data with analogous accuracy but reduced computational costs, thus providing surrogate models rather than enhanced data-enabled physics insights. Although ML-based surrogate models are useful to overcome current limitations associated with the high computational costs of CFD models, using ML to unveil processes from experimental data or enhance modeling capabilities is deemed a potential research direction to pursue. In this letter, we discuss recent achievements in the realm of ML modeling of wind turbine wakes and operations, along with new promising research strategies. 

Motivated by the computational difficulties incurred by popular deep learning algorithms for the generative modeling of temporal densities, we propose a cheap alternative that requires minimal hyperparameter tuning and scales favorably to high-dimensional problems. In particular, we use a projection-based optimal transport solver [Meng et al.,Advances in Neural Information Processing Systems (Curran Associates, 2019), Vol. 32] to join successive samples and, subsequently, use transport splines (Chewi et al., 2020) to interpolate the evolving density. When the sampling frequency is sufficiently high, the optimal maps are close to the identity and are, thus, computationally efficient to compute. Moreover, the training process is highly parallelizable as all optimal maps are independent and can, thus, be learned simultaneously. Finally, the approach is based solely on numerical linear algebra rather than minimizing a nonconvex objective function, allowing us to easily analyze and control the algorithm. We present several numerical experiments on both synthetic and real-world datasets to demonstrate the efficiency of our method. In particular, these experiments show that the proposed approach is highly competitive compared with state-of-the-art normalizing flows conditioned on time across a wide range of dimensionalities. 

Abstract The power performance and the wind velocity field of an onshore wind farm are predicted with machine learning models and the pseudo‐2D RANS model, then assessed against SCADA data. The wind farm under investigation is one of the sites involved with the American WAKE experimeNt (AWAKEN). The performed simulations enable predictions of the power capture at the farm and turbine levels while providing insights into the effects on power capture associated with wake interactions that operating upstream turbines induce, as well as the variability caused by atmospheric stability. The machine learning models show improved accuracy compared to the pseudo‐2D RANS model in the predictions of turbine power capture and farm power capture with roughly half the normalized error. The machine learning models also entail lower computational costs upon training. Further, the machine learning models provide predictions of the wind turbulence intensity at the turbine level for different wind and atmospheric conditions with very good accuracy, which is difficult to achieve through RANS modeling. Additionally, farm‐to‐farm interactions are noted, with adverse impacts on power predictions from both models. 

Wind turbine wakes are responsible for power losses and added fatigue loads of wind turbines. Providing capabilities to predict accurately wind-turbine wakes for different atmospheric conditions and turbine settings with low computational requirements is crucial for the optimization of wind-farm layout, and for improving wind-turbine controls aiming to increase annual energy production (AEP) and reduce the levelized cost of energy (LCOE) for wind power plants. In this work, wake measurements collected with a scanning Doppler wind Li- DAR for broad ranges of the atmospheric static stability regime and incoming wind speed are processed through K-means clustering. For computational feasibility, the cluster analysis is performed on a low-dimensional embedding of the collected data, which is obtained through proper orthogonal decomposition (POD). After data compression, we perform K-means of the POD modes to identify cluster centers and corresponding members from the LiDAR data. The different cluster centers allow us to visualize wake variability over ranges of atmospheric, wind, and turbine parameters. The results show that accurate mapping of the wake variability can be achieved with K-means clustering, which represents an initial step to develop data-driven wake models for accurate and low-computational-cost simulations of wind farms. 

Abstract Flow modifications induced by wind turbine rotors on the incoming atmospheric boundary layer (ABL), such as blockage and speedups, can be important factors affecting the power performance and annual energy production (AEP) of a wind farm. Further, these rotor‐induced effects on the incoming ABL can vary significantly with the characteristics of the incoming wind, such as wind shear, veer, and turbulence intensity, and turbine operative conditions. To better characterize the complex flow physics underpinning the interaction between turbine rotors and the ABL, a field campaign was performed by deploying profiling wind LiDARs both before and after the construction of an onshore wind turbine array. Considering that the magnitude of these rotor‐induced flow modifications represents a small percentage of the incoming wind speed ( ), high accuracy needs to be achieved for the analysis of the experimental data and generation of flow predictions. Further, flow distortions induced by the site topography and effects of the local climatology need to be quantified and differentiated from those induced by wind turbine rotors. To this aim, a suite of statistical and machine learning models, such as k‐means cluster analysis coupled with random forest predictions, are used to quantify and predict flow modifications for different wind and atmospheric conditions. The experimental results show that wind velocity reductions of up to 3% can be observed at an upstream distance of 1.5 rotor diameter from the leading wind turbine rotor, with more significant effects occurring for larger positive wind shear. For more complex wind conditions, such as negative shear and low‐level jet, the rotor induction becomes highly complex entailing either velocity reductions (down to 9%) below hub height and velocity increases (up to 3%) above hub height. The effects of the rotor induction on the incoming wind velocity field seem to be already roughly negligible at an upstream distance of three rotor diameters. The results from this field experiment will inform models to simulate wind‐turbine and wind‐farm operations with improved accuracy for flow predictions in the proximity of the rotor area, which will be instrumental for more accurate quantification of wind farm blockage and relative effects on AEP.