Presentations

Abstract

In a review of annual energy production ( AEP) assessments, comparing model predictions for turbine interaction losses (wakes and blockage), DNV has identified a gap between the predictions from its well-validated high-fidelity CFD model (HiFi CFD) and commonly used engineering (CUE) models, with the latter tending to predict lower losses. The benchmark is showing a trend whereby the size of the gap increases as the magnitude of the turbine interaction losses increases. Discrepancies of up to 5% are seen for the largest projects, between HiFi CFD and CUE models. The HiFi CFD validates more consistently than CUE models across a range of project sizes [1] and by virtue of including more of the controlling physics and resolving those physics to turbine scale, the HiFi CFD is expected to generalize better to future project scales. Thus, there is an elevated risk of overestimated AEP for future mega projects, if assessed only with CUE models. While HiFi CFD is a solution to mitigate the identified risk, rapid modelling approaches are still required to quantify turbine interaction losses at early stages of project development.  A rapid modelling alternative to CUE models is CFD.ML (version 2) [2]. This machine learning model, trained on thousands of HiFi CFD flow cases, can be used to derive a blockage correction for a CUE model for wakes, or to directly predict the total turbine interaction losses.  In addition to the benchmark results for CUE models, this contribution will present results from the evaluation of updated rapid modelling approaches involving WindFarmer CFD.ML and their ability to close the identified gap. The benchmark data set consists of over 50 wind farms design scenarios where HiFi CFD results were available. Approximately half of these use DNV’s WRF-to-CFD approach and were conducted during the last 3 years. The wind farms span the globe, onshore and offshore, forming the worlds’ largest database of HiFi CFD turbine interaction predictions, from which turbine interaction AEP losses may be derived. The projects are typically large, or complex, where concerns about turbine interaction modelling incentivized a high-fidelity wind farm flow analysis to reduce uncertainties. The database includes projects at scales significantly beyond those operating today, including farm clusters of > 1000 turbines with turbine technology up to 23 MW. The presentation will describe how HiFi CFD model results can help define an appropriate treatment of uncertainties at the unvalidatable mega project scales of tomorrow. Based on this we will discuss how risk can be mitigated by taking a multi-fidelity approach to turbine interaction modelling, and how uncertainties can be reduced earlier in the project lifecycle with new model combinations.  References 1.  C. Montavon et al, Cluster wakes and their effect on a wind farm annual energy production, 2024, DNV WhitePaper. 2. T. Levick et al, Boundary layer educated long range wake estimates from WindFarmer CFD.ML, 2024, WindEurope Tech Workshop, Dublin