A combination of weather clustering and data-driven methods for intraday PV power production forecasting

Abstract

PV systems are becoming the primary solution for meeting future electricity demands as conventional power generation declines globally. However, integrating PV systems into distribution networks presents new challenges in maintaining continuous service during supply and demand fluctuations. To address these challenges, a new intraday PV production forecasting methodology is proposed for predicting production 1 to 3 hours ahead. The methodology involves data quality assessment, weather clustering, and machine learning modeling. The proposed methodology demonstrated errors below the 5% theoretical threshold.