基于交叉注意力融合的短时光伏功率预测OA

To improve the accuracy of short-term photovoltaic(PV)power forecasting for solar PV sys-tems under complex meteorological conditions,a multimodal short-term forecasting model named ViT-Informer is proposed.The Vision Transformer(ViT)and Informer are utilized to extract image fea-tures and temporal features,respectively,followed by a cross-attention mechanism to achieve bimodal feature fusion,capturing the key characteristics causing power fluctuations,thereby improving the ac-curacy of photovoltaic power prediction.Experiments conducted on a self-built small-scale PV system integrated with an all-sky imaging platform demonstrate that the attention mechanism can effectively extract critical features while suppressing redundant information.For 15-minute-ahead forecasting,compared with single-modality models,the proposed fusion model improves the prediction skill score(Fs)by 9.5％～28.2％,and reduces the normalized mean absolute error(NMAE)and normalized root mean square error(NRMSE)by up to 9.8％and 5.4％,respectively.