Posters

Abstract

The objective of this study is to develop an efficient turbine specific wake induced wind field parameterization and to achieve surrogate modeling for loads (ultimate and fatigue) and power prediction in an arbitrary wind farm layout. The current design approach for wake induced load analysis just consists in using a kind of equivalent turbulence intensity that would cause the same damage as the wake induced wind field. But this low fidelity approach does not take into account the specific shape of the wake velocity field and may lead to inappropriate (conservative or not) load predictions especially for rotor operating in partial-wake conditions. On the other hand, the major problem with high fidelity approaches is the huge computational burden. The objective is hence to implement an efficient « computing route » to predict wake induced ultimate and fatigue loads plus the power for arbitrary layout with current research engineering software including partial wake conditions, added turbulence and velocity deficit phenomena. This parameterization must be compatible with current research engineering software (or model chain) and be reasonable in terms of number of parameters, especially if the model chain involves fully coupled FEA software. This limitation is desirable for two reasons: i/ some surrogate modeling techniques are limited in terms of number of dimensions (polynomial chaos and kriging) ii/ to avoid the curse of dimensionality in terms of design of experiment (DoE) when considering fully coupled FEA simulation cost. The wind farm parametrization proposed here considers a "mean wind speed profile-oriented parametrization". Basically, it involves finding a suitable parametrization of the y-z plane mean wind velocity field (from Farmshadow static wake software of IFPEN or any other sources) at the wind turbine location with specific functions. Then, these functions are used to modify turbulent wind field generated by any turbulence box generator. This approach may achieve a fidelity level that would be in-between the proposed approach in the current design standards and the dynamic approaches such as the DWM of DTU. After a brief description of the model function and the associated “wake parameters” needed to fit a y-z plane mean wind velocity field inside the farm, the fitting process is presented. Then, to get a catalog of wind field profiles specific to the Teesside wind farm, a Farmshadow DoE is processed. Based on this catalog and the fitting process, the joint distributions of the wake parameters can be used to build a relevant DoE for DeepLines WindTM simulations. These simulations are then used to train Gaussian process surrogate models mapping the wake parameters and some quantities of interest (Damage Equivalent Load, Ultimate Load or Electrical Power). Ultimately, a prediction of these quantities with the surrogates can be quickly obtained for any set of wake parameters and a Global Sensitivity Analysis (GSA) may also be conducted.