Effective Input Dataset Identification Methodology for Accurate Prediction of Local PV Power Production

Abstract

Local photovoltaic (PV) systems are playing a considerable role globally as a power resource and constituent element of the smart grid. Nevertheless, PVs may cause significant problems to the electric grid. This is due to the high variability of the PV power that is instigated by intermittent environmental conditions. Accurate prediction of PV power is very important to operate power grids containing high penetration of PVs. Most prior approaches have focused on forecasting the collective quantity of solar power generation at national or regional scale and disregarded the local PVs that are installed mainly for local electric supply. This paper devises an effective input dataset identification methodology (IDIM) to find the most significant and non-repetitive input variables for accurate prediction of the power production of local PVs. In the devised methodology, the Binary Genetic Algorithm (BGA) is used for the input variable identification and Support Vector Regression (SVR) is employed for evaluating the fitness of the input datasets. The devised methodology is implemented and validated based on actual local PVs (building rooftop PVs) located in the Otaniemi area of Espoo, Finland. The results are compared with those obtained by conventional counterparts and manifest outperformed performances.