Distributed photovoltaic ultra-short-term power prediction method based on short-term cloud image pixel characterization

The embodiment of the invention discloses a distributed photovoltaic ultra-short-term power prediction method based on short-term cloud image pixel representation. The distributed photovoltaic ultra-short-term power prediction method is based on short-term cloud image pixel representation. The method comprises the following steps: acquiring cloud picture data; processing and predicting the cloud picture data to obtain a prediction result; performing feature region positioning on the prediction result to obtain a feature descriptor; fusing the feature descriptors with influence factors to obtain a fusion result; inputting the fusion result into a radial basis function neural network for surface radiation prediction to obtain a surface radiation prediction result; and performing photoelectric conversion on the surface radiation prediction result to obtain a power prediction result. By implementing the method provided by the embodiment of the invention, the distributed photovoltaic ultra-short-term power can be