Weather classification is an indispensable preprocessing step in photovoltaic (PV) power prediction. A new two-modal weather classification methods based on PV power clustering was proposed to finely depict the uncertainty of PV power output. Both PV power data and meteorological data were considered for weather classification, providing a novel and effective path for PV power prediction. In addition, data fusion technology was used to extract relevant information from both numeric weather prediction (NWP) data and measured meteorological data to help for weather classification. This approach reduces the model’s reliance on the accuracy of forecasted meteorological indicators and improve the robustness of the model. Experiments based on data from a PV power station in Jilin demonstrated the rationality of the proposed weather classification method. Combining the PV power probability prediction with the proposed weather classifier resulted in prediction interval coverage probabilities closer to the preassigned confidence level and narrower mean prediction interval width.
