Research on Photovoltaic Interval Prediction Method Based on Meteorological Correction and Transformer

This study proposes a method to predict photovoltaic (PV) power generation intervals. It combines meteorological correction and the Transformer model. Historical data were preprocessed and corrected using linear regression to reduce forecast errors. The Transformer model captured spatiotemporal features, and Monte Carlo (MC) dropout was used for uncertainty estimation. Training and cross-validation were performed using historical data, and the model provided accurate and reliable PV output prediction intervals, beneficial for power scheduling and management.