Development of an automated multi model and multi-parameter variation tool for estimating wake losses and their uncertainties

In wind farm yield estimation, wake losses and associated uncertainties play a centralrole. This work presents an automated tool for estimating wind farm yields and relateduncertainties by varying different analytical and computationally efficient wake models.Additionally, the wake models are extended by varying empirical parameters, such as thewake expansion parameter. To quantify the uncertainty of the estimated wake losses, thespread of model and parameter variations is utilized. The wake models and their simulationsfor estimating wind farm yields are initialized and executed using the Python-basedsoftware PyWake. The tool is then used to perform simulations to estimate the AnnualEnergy Production (AEP) at the Langenburg onshore test site. To test and validate thetool, the estimated AEP is compared with the data from the Supervisory Control andData Acquisition (SCADA) system of the test site. Since the Langenburg onshore test siteis situated in complex terrain and within a forest, these factors are considered through awind field generated with the PyWAsP software. This introduces the challenge that thesefactors, in addition to wake losses, induce further uncertainty. To evaluate this additionaluncertainty, two simulations are conducted for the test site: one that does not account forterrain and surface roughness, and another that includes these influences through thegenerated wind field with PyWAsP. Both simulation results show a systematic overestimationof AEP for all wind turbines, with one exception. It is observed that consideringterrain and surface roughness slightly improves the AEP estimation for most turbines.This systematic overestimation aligns with expectations, as the presented tool only considerswake losses and does not cover all loss types, such as blockage effects or yawmisalignment. Furthermore, the uncertainty associated with the AEP estimated throughmodel and parameter variation is quantified using the Standard Deviation (STD). To evaluatethe accuracy of the STD as a predictor of actual uncertainty, it is compared with theRoot Mean Square Deviation (RMSD) of the model outputs against the SCADA data. Itwas found that the prediction of uncertainty does not perform equally well for everyturbine. Additionally, the validity of the STD cannot be assured. The results indicate astrong need to test the developed tool on additional wind farms and to expand its capabilitiesto include more loss types and models.