Posters

Abstract

This study focuses on enhancing the predictability of monthly potential capacity factor for a wind energy complex over a 7-month horizon compared to the pre-construction P50 estimate. Recognizing the critical implications for the wind energy sector, ranging from economic viability assessments to operational improvements, accurate forecasting remains a substantial challenge. Six models were evaluated, including Seasonal Naive, SARIMA, ETS, and Prophet—relying solely on observed generation and wind speed history. Multiple linear regression and Random Forest models incorporated meteorological predictions. Among these, the Random Forest model emerged as the most robust, achieving the lowest error metrics: MAE equal to 0.058, RMSE equal to 0.07, and MAPE equal to 12.85%. Additionally, the sMAPE equal to 11.96% was attained by the simpler Seasonal Naive model. Remarkably, the Random Forest model demonstrated an average reduction of approximately 45% in error metrics compared to the pre-construction P50 model. These findings underscore the potential of advanced modeling techniques, particularly Random Forest, in significantly improving the accuracy of wind energy production forecasts. The implications extend to maintenance planning, and overall energy portfolio management, reinforcing the pivotal role of precise wind energy predictions in the renewable energy landscape.