Answers are grounded in this paper and cite their sources.
Urban wind energy applications are characterized by low wind speeds, variable wind directions, and turbulent flow conditions that influence small-scale wind turbine performance. Although machine learning methods have been widely investigated for wind speed forecasting, the implications of prediction errors for turbine energy production and carbon emission reduction remain less extensively examined. This study evaluates the effects of machine-learning-based wind-speed predictions on the estimated energy production and carbon emission reduction potential of two commercial 10 kW horizontal-axis (HAWT) and vertical-axis (VAWT) wind turbines under inland urban wind conditions. Five prediction models, namely Linear Regression, Support Vector Machine, Random Forest, XGBoost, and Long Short-Term Memory, were evaluated using hourly meteorological measurements collected in Karaman, Türkiye, over one year. Turbine power output was estimated from measured and predicted wind speeds using power curves with corrections for local air density. Among the evaluated models, XGBoost achieved the lowest root mean square error (RMSE) of 0.5957 m/s and the highest coefficient of determination (R2) of 0.7961 on the chronological test dataset. However, the predicted wind speeds showed 15.1% lower variability than the measurements, underrepresenting higher-wind-speed events. Based on measured wind speeds, the equivalent annual energy production was estimated at 0.6559 MWh for the HAWT and 0.9622 MWh for the VAWT, corresponding to capacity factors of 0.75% and 1.10%, respectively. Using XGBoost predictions reduced these estimates by 30.7% and 24.3%, while the corresponding net annual CO2 emission reductions decreased from 0.4041 to 0.2801 tCO2 for the HAWT and from 0.5928 to 0.4487 tCO2 for the VAWT. Variance scaling and quantile mapping reduced the magnitude of the energy estimation errors, although the wind-speed prediction errors increased. Under the assumed cost parameters, the estimated levelized costs of energy exceeded 6 USD/kWh, indicating limited economic feasibility for both turbines at the investigated site. The results indicate that wind-speed prediction accuracy alone does not fully characterize the reliability of subsequent energy and carbon assessments, particularly under low-wind conditions.
No takes yet. Share an insight, caveat, or question.
Gulhan et al. (2026) studied this question.