Wind farm parameterization and turbulent wind box generation - IFPEN - IFP Energies nouvelles Access content directly
Reports Year : 2022

Wind farm parameterization and turbulent wind box generation

Abstract

The present report addresses Hiperwind Deliverable 3.1: wind farm parameterization and turbulent wind box generation. This includes two separate scopes: 1) studying the effect of wakes on wind turbine loads - and more specifically how to describe (parameterize) the wake effects in terms of quantities suitable for surrogate model inputs, and 2) turbulent wind field generation for special scenarios including transient events (extreme wind gusts, ramps and direction changes) and situations with non-Gaussian statistics of the wind field. The common for both topics is that they represent aspects of the inflow of wind turbines in wind farms that are a major factor affecting the wind turbine loads. Thus, the main purpose of this study is to develop methods and tools for more efficient and accurate modelling of such scenarios, and to better understand the uncertainties involved and the relative significance of the different inputs on wind turbine loads and power. Summary: wind parameterization in a wind farm We studied two wind farm parameterization methods that are suitable for two different workflows for wind farm load assessment. In the first method developed by DTU, wind turbine loads are simulated using the Hawc2 aeroelastic tool, with the Dynamic Wake Meandering (DWM) model used to include a dynamic simulation of incoming wakes. In order to define appropriate inputs for a surrogate model that can replace the Hawc2+DWM simulations, the wake effects are described in terms of variables related to the geometric layout of the wind farm such as relative wind turbine positions. Several parameterization approaches were tested within this workflow: encoding based on the overlap between the wake deficit and the rotor, on the relative upwind turbine positions, and on an autoencoder model that applies dimensionality reduction on the relative position encoding. The encoding obtained in this way served as an input to a regression model based on Feedforward Neural Networks. The second parameterization method developed by IFPEN is based on using shape functions to describe the wake deficit properties. First, a “wake parameterization” surrogate model is trained that uses a few parameters to describe a mixture of Gaussian shape functions that define the averaged properties of the wake deficit under a broad range of waked inflow conditions. This surrogate is trained on simulations with the FarmShadow tool. Then, the predicted wake deficit is superimposed on a turbulent wind field and fed into load simulations with Deeplines Wind. Finally, a second surrogate model is trained that maps the wake parameterization surrogate outputs to the outputs of load simulations. Combining the two models, the loads can be predicted for a wind turbine in an arbitrary wind farm layout. Sensitivity analysis in terms of Sobol indices was carried out with both parameterization approaches, and the results from the load assessment and sensitivity analysis were compared. The outcome of the wake parameterization studies led to the following conclusions: - It was shown that it is feasible to parameterize the wind farm layout in a way that enables the construction of surrogate models that predict loads and power outputs of individual turbines in arbitrary wind farm layouts; - The sensitivity analysis showed that the ambient wind speed is the governing factor for loads and power production. The wind direction (which determines the strength of the wake effects) and the ambient turbulence are also of high importance. For wind speeds just below rated where the turbine thrust is highest, the wind direction has higher importance than the turbulence, while at low wind speeds the turbulence has more significant effect. For all methods that were studied, the wind shear was found to be of small significance. - The performance of the surrogates is affected by the significance of wake effects: for signals with less wake impact, such as the extremes, the performance is consistently higher Summary: uncertainty reduction in turbulent wind box generation The second inflow aspect we considered is the turbulent field that is used as input to load simulations. The focus is on situations that deviate from the normal operation under stationary conditions where the standard turbulence generation approaches provide sufficient quality of results. The special scenarios considered were transient events (extreme wind gusts, ramps and direction changes), and situations with highly non-Gaussian field statistics. For the purpose of this investigation, an open Python-based turbulence field generation tool was created within the Hipersim software package. The tool generates turbulence fields that can serve as inputs to aeroelastic load simulations, and can embed measured wind time series as constraints in the wind field. The capabilities were further expanded by introducing the possibility to generate non-Gaussian fields with a predefined skewness and kurtosis. Using the constrained simulation approach, the load response of the DTU10MW turbine was simulated under a set of transient events that were previously obtained from measurements as part of Hiperwind Deliverable 2.3. The load results from these events were compared to simulations with the transient design load cases defined in the IEC61400-1 standard. In addition, the effect of introducing non-Gaussian statistics was assessed by carrying out pairs of load simulations with the same turbulence seeds but different skewness and kurtosis. The following conclusions were drawn: - It was verified that the Hipersim/Turbgen tool produces turbulence boxes with the correct spectrum and coherence properties as prescribed with the Mann turbulence spectrum; - The constrained simulation functionality of Hipersim results in correct reproduction of the target time series in the turbulence boxes; - Overall, the loads obtained using simulations with measured transient events were in most cases less than the loads resulting from the synthetic events prescribed by the IEC standards. However, there are also cases where the constrained turbulence loads were either similar or exceeded the IEC 61400 loads. - The synthetic conditions from the IEC standard tend to produce a different relationship between wind speed and load magnitude, compared to the measured time series simulations. The highest loads from IEC events tended to be concentrated near rated wind conditions, while the highest constrained turbulence loads were more widely distributed over more wind speeds. - A non-Gaussian turbulence box generator was developed that produces turbulence boxes with the Veers model and arbitrary skewness and kurtosis. It was verified that the resulting turbulence fields attain the target statistical moments, while retaining the correct spectrum and coherence. - The present method of non-Gaussian turbulence generation is based on statistical rather than physical considerations. Although the results are statistically consistent, there is a chance that they are unphysical as e.g. major intermittent wind direction reversals. Increasing the skewness and kurtosis of a wind time series amplifies the chance of such non-physical events. - For most load channels, the effect of non-Gaussian wind statistics was mild, such as minor changes in the design driving load and a shift in the wind speed where the dominant load events occur. An exception were the tower top yaw moment extremes, where changes in the skewness and kurtosis led to a significant increase in the loads. Research significance The research activities within this study provide solutions for more accurate and more computationally efficient assessment of loads under challenging scenarios such as wakes, transients, and unusual turbulence structures that have non-Gaussian statistics. This provides the tools for more efficient wind turbine and wind farm design process, and opens possibilities for uncertainty-aware design and optimization. In addition, the outcomes of this work increase our understanding of the significance of wind turbine wakes, ambient wind conditions and transients on wind turbine loads. This provides insights into what type of scenarios need increased focus in future design and planning activities, and potentially will influence the future versions of wind turbine design standards.
Fichier principal
Vignette du fichier
HIPERWIND_D3.1.pdf (10.73 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-04033050 , version 1 (16-03-2023)

Identifiers

  • HAL Id : hal-04033050 , version 1

Cite

Michael Mc William, Nicolas Bonfils, Nikolay Dimitrov, Suguang Dou. Wind farm parameterization and turbulent wind box generation. DTU; IFPEN. 2022, pp.Deliverable n° D3.1. ⟨hal-04033050⟩

Collections

IFP LARA
102 View
30 Download

Share

Gmail Facebook X LinkedIn More