Analysis of engineering wake model validation and calibration with historical data from OWEZ wind farm

Abstract

Wind turbines are often placed together in wind farms for economic considerations. This causes wake interactions between turbines, resulting in significant power losses. Models that predict these wake losses are critical for estimating wind farm power output and developing strategies to mitigate the wake effect, such as wind farm control. For these applications, engineering wake models are favoured for their computational efficiency. Hence, the validation and improvement of these models is an ongoing area of research. Currently, consensus on the accuracy of engineering wake models is absent in the literature. Existing studies employ varying validation strategies that impact the perceived model accuracy. Furthermore, proposed model improvements often lack quantitative evaluation, limiting the generalisability of the results. Additionally, the potential benefits of calibrating wake model parameters are recognised, yet research on calibration methods and the impact thereof is limited.

This thesis addresses this scientific gap by proposing a holistic framework for the validation and calibration of engineering wake models. The framework combines best practices from literature. First, it accounts for wind direction uncertainty in historical wind farm data. Additionally, it corrects model inputs by including heterogeneous inflow wind speeds. Finally, it offers a methodology for parameter calibration to improve the model's accuracy using historical wind farm data. The overarching framework employs both quantitative and qualitative validation methods to mitigate the impact of experiment design and enable a thorough evaluation of model improvements. The effectiveness of this framework is demonstrated through a case study with SCADA data from OWEZ wind farm and four engineering wake models from the popular control-oriented wake modelling tool FLORIS.

Results show that wind direction uncertainty in SCADA d|ata must be included when validating wake models for specific wind directions or sectors. Additionally, incorporating heterogeneous inflow wind speeds reduced the absolute turbine error by up to 20%. Furthermore, it is demonstrated that calibrating model parameters significantly improves model accuracy. The resulting error reductions reach up to 92% for individual turbines and 65% at farm-level, i.e., for all turbines collectively. Furthermore, results revealed that while the performance of the different models converges post-calibration, differences persist in various scenarios with numerous wake interactions. In these cases, the CC and TurbOPark models outperform the Jensen and GCH models.

Through this holistic framework and the demonstrated potential of model parameter calibration, a path forward is paved for further model improvement in a systematic and quantitative manner.