Turbine loading and wake model uncertainty
Résumé
The present report addresses HIPERWIND Deliverable 3.2: “Turbine loading and wake model uncertainty”. The following work provides new results to estimate the uncertainty of analytical wake models. Also known as “engineering wake models”, such tools are widely used to predict the production of a wind farm at a given site during the pre-design stage. Their main advantage is a low computational cost, obtained through simplified formulations to describe the behaviour of wakes. Nevertheless, the assumptions considered may lead to limitations. Consequently, such models will be compared in section 2 to medium-fidelity models with Dynamic Wake Meandering (DWM) and high-fidelity models involving Large Eddy Scale (LES) with actuator methods. Wake conditions are not only important for electricity production but also for the load on the turbines, specifically for fatigue loads on blades and towers. The fatigue life of such components may be strongly influenced by the added turbulence, and spatial deviation induced e.g., by yaw misalignment of upwind turbines or meandering. The results of this report are however only limited to the wind conditions seen at different locations within a wind farm which will define the input aerodynamic loading for future reliability-based designs of offshore wind turbines (WT).
These results are documented for the two offshore case studies of HIPERWIND: the 2.3 MW WT on a monopile within the Teesside (United Kingdom) wind farm, and the IEA15MW WT on the UMaine semi-submersible floater in the South Brittany site (France). The latest is a modified version of the original design which was proposed by NREL and UMaine [1] [2][3] [4].
For the floating case, a new wake tool has been developed which takes into account the static position of the floater due to constant mean wind forces on the WT. Section 3 presents the main difference in wake predictions (mean wind speed and turbulence) when compared to a fixed case. It is shown that the influence is small and mainly due to the inclination of the floater. Annual Electricity Production (AEP) is computed for both fixed and floating cases. The conclusion is that the main difference is coming from the rotor tilt of a free WT and that the wake modification is negligible.
The uncertainty of engineering methodology for wake modelling is investigated in section 4. Firstly, a qualitative comparison to high-fidelity LES simulations is conducted for selected representative configurations of the ambient wind conditions. Differences in wake behaviours are noticed, depending on the atmospheric stability and the static inclined position of the floater. If the engineering approach provides fairly good results in neutral conditions, more discrepancies are observed in stable ones (underestimation of speed deficit). Concerning the inclined position of the floater, the vertical wake deflection seems well predicted, but a wider and stronger wake is observed in LES simulations downstream of the first wind turbine. The turbulence intensity shows also good predictions. A slight and overall overestimation underlines the need for further investigations to improve such models. Secondly, a quantitative estimation of uncertainty has been computed with a Kriging approach, considering Dynamic Wake Meandering models as a reference. Once more, the overall differences for wake deficit remain small, while higher discrepancies are evaluated for turbulence intensities. Furthermore, an important uncertainty is identified when the wind turbine is directly downstream of several others, highlighting the need for improved superposition models.
Finally, section 5 is devoted to the uncertainty propagation, from ambient wind conditions to parameters defining the local wind conditions which should be used in a Reliability-Based Design (RBD) of the offshore WT composing a farm. A metric based on Maximum Mean Discrepancy (MMD) is defined to quantify the difference between ambient mean wind speed plus turbulence intensity and the corresponding quantities modified by the wake. The latter quantities are integrated over the rotor, at each WT of the farm. RBD may require huge computational costs which may be intractable in an industrial context. Such costs are due to the cost of a single multiphysics simulation plus the high number of simulations required to sample the low target failure probability in standard (e.g., 1e-4 for the normal safety class in [5]). To alleviate this problem, a clustering strategy is proposed to group the WT into a reduced number of sets, according to the similarity in the local wind uncertainty metric. A demonstration is given for the Teesside case study with only 5 clusters so that only 5 WT of the farms provide an RBD analysis for the whole set of WT.