Evaluation of an offshore wind farm computational fluid dynamics model against operational site data

Tools

Richmond, M. and Antoniadis, A. and Wang, L. and Kolios, A. and Al-Sanad, S. and Parol, J. (2019) Evaluation of an offshore wind farm computational fluid dynamics model against operational site data. Ocean Engineering, 193. pp. 1-12. 106579. ISSN 0029-8018 (https://doi.org/10.1016/j.oceaneng.2019.106579)

[thumbnail of Richmond-etal-OE2019-Evaluation-of-an-offshore-wind-farm-computational-fluid-dynamics-model]
Preview
 Text. Filename: Richmond_etal_OE2019_Evaluation_of_an_offshore_wind_farm_computational_fluid_dynamics_model.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (6MB)| Preview

Abstract

Modelling wind turbine wake effects at a range of wind speeds and directions with actuator disk (AD) models can provide insight but also be challenging. With any model it is important to quantify the level of error, but this can also present a challenge when comparing a steady-state model to measurement data with scatter. This paper models wind flow in a wind farm at a range of wind speeds and directions using an AD implementation. The results from these models are compared to data collected from the actual farm being modelled. An extensive comparison is conducted, constituted from 35 cases where two turbulence models, the standard k-ε and k-ω SST are evaluated. The steps taken in building the models as well as processes for comparing the AD computational fluid dynamics (CFD) results to real-world data using the regression models of ensemble bagging and Gaussian process are outlined. Turbine performance data and boundary conditions are determined using the site data. Modifications to an existing opensource AD code are shown so that the predetermined turbine performance can be implemented into the CFD model. Steady state solutions are obtained with the OpenFOAM CFD solver. Results are compared in terms of velocity deficit at the measurement locations. Using the standard k-ε model, a mean absolute error for all cases together of roughly 8% can be achieved, but this error changes for different directions and methods of evaluating it.

ORCID iDs

Richmond, M., Antoniadis, A., Wang, L., Kolios, A. ORCID logoORCID: https://orcid.org/0000-0001-6711-641X, Al-Sanad, S. and Parol, J.;
CORE (COnnecting REpositories)