Presentations

Abstract

Mainstream carried-out an investigation into the performance of wind flow and wake models at 10 operating wind farms using SCADA data, to understand the strengths and weaknesses of each model. For wind flow modelling, 5 models were analysed: linear (Openwind WindMap), mesoscale (Vortex FARM), CFD RANS (WindSim) Neutral and Coupled (with and without temperature equations activated, following "A new meso-microscale coupled modelling framework for wind resource assessment: A validation study" from Duran et al. [https://doi.org/10.1016/j.renene.2020.06.074] ). Two metrics were retained for assessing the accuracy of each wind flow model compared to operational production data: * The coefficient of determination between the individual operational turbine AEP and that modelled; as operational AEP inherently includes the wake losses, a second metric was used * The relative error between the operational AEP at a couple of sample turbines and the modelled AEP. Sample turbines were selected to have minimum modelled wake losses (<3%) and to be geographically representative of the site. For wake modelling, initially 5 neutral wind flow models were compared to operational data (for a total of 20 pairs of wind turbines analysed and 36 wake cases): Eddy Viscosity Deep Array Wake Model (EV DAWM, with and without near wake filter), WakeBlaster and TurbOPark (for two expansion calibration parameters) - all models as implemented in latest version of Openwind. The results of this comparison will be presented first - note that TurbOPark is a wake model designed for offshore wind farms while the wind farms studied here are onshore; some adjustments to the wake calibration expansion parameter (A) are suggested consequently. But further investigation highlighted the importance of considering atmospheric stability in wake modelling, as highlighted in previous papers from the industry. Further binning of the operational power deficit per turbulence bin and Pasquill Class (using Obukhov Length derived from mesoscale model) emphasised the importance of atmospheric stability on wake modelling, even for the same turbulence values, and shows that even the most representative model used here (WakeBlaster, with atmospheric stability enabled) tends to underestimate wake propagation in very stable atmosphere. It also shows the limits of neutral wake modelling (EV DAWM and TurbOPark in that case). This work might suggest some leads to improve the current wind flow and wake modelling tools in the industry for more accurate AEP estimates.