Wind farm performance depends critically on how turbine wakes—the disturbed flow regions downstream of each turbine—are modeled during the design phase. Poor wake representation leads to overestimated energy production forecasts, reduced revenues, and unexpected structural stresses on downstream machines. This research validates computational approaches for predicting wake effects and optimizing turbine placement to maximize annual energy production.
The study evaluated three modeling strategies: an artificial neural network trained on high-fidelity Reynolds-averaged Navier–Stokes (RANS) computational fluid dynamics simulations, and two physics-based engineering models using TurbOPark and super-Gaussian wake formulations. Engineers systematically tested these approaches across varying numbers of turbines and inter-turbine spacing configurations.
Key findings show the ANN surrogate achieved superior prediction accuracy with the lowest root-mean-square error and mean absolute percentage error in flow field estimation. However, this accuracy advantage did not automatically translate to better optimized layouts; the recalibrated TurbOPark model actually produced designs with higher validated annual energy production when benchmarked against RANS baseline data. This counterintuitive result highlights that optimization landscape complexity—influenced by wake interactions, boundary conditions, and local terrain—sometimes favors models that balance predictive accuracy with computational tractability.
Additionally, the research found that incorporating blockage effects (where wind slows before reaching the farm) increased computational demands without meaningfully improving results. This suggests practitioners can streamline calculations by excluding blockage modeling in routine optimization studies.
These findings have practical implications for wind project developers. While machine learning surrogates offer advantages in speed and prediction fidelity for research applications, traditional engineering models remain competitive for real-world optimization when properly recalibrated. A hybrid strategy—using ANN models to explore design space broadly while validating final layouts with TurbOPark-based optimization—may offer the best balance of accuracy, speed, and confidence for operational decision-making.
