Revenue discrepancies in wind power: how data imputation methods influence yield estimates
Аннотация
Transitioning to sustainable energy demands precise resource assessment tools, accurate forecasting, and informed decision-making. In wind energy projects, incomplete datasets pose a major challenge, leading to biased assessments and uncertainty. While many imputation techniques have been suggested, their economic impacts are rarely studied in depth. This study compares seven imputation methods, including statistical, machine learning (ML), and deep learning (DL) approaches to fill missing wind speed data. The simulation data with missingness rates from 5% to 65% were evaluated using the mean absolute error (MAE), the normalized root mean square error (NRMSE) and $$R^2$$. Beyond statistical metrics, the analysis includes revenue impacts through turbine power curves and electricity prices to estimate energy from imputed series and compare with the reference. The results show that ML methods outperform traditional techniques at low missing data (5%), while DL performs best under high data loss conditions (65%). Revenue deviations are minimal at low missing rates, but at high data loss ML underestimates revenue by $$\approx 16\%$$, whereas DL limits this to $$\approx 7\%$$. This study proposes a new evaluation framework, offering a decision support system for the wind energy sector to improve risk management and planning with incomplete data, enhancing understanding of the link between data quality, economic estimates, and decision-making in renewable energy projects.