Improving Forecast Reliability for Geographically Distributed Photovoltaic Generations

Photovoltaic (PV) generation is potentially uncertain. Probabilistic PV forecasting methods have been proposed with prediction intervals (PIs) to evaluate the uncertainty quantitively. However, few studies applied PIs geographically distributed PVs in a specific area. In this study, two-step probabilistic forecast scheme for forecasting. Each step of adopts ensemble based on three different machine-learning methods. When individual forecasted, utilizes surrounding PVs’ past data train model. case was compared conventional non-multistep The improved reliability and deterministic results through 30 days continuous operation real Japan.