In view of multidimensional meteorological data with strong coupling

on the basis of obtaining valid data in photovoltaic multi-sensor system（ MUSS）

a data mining photovoltaic（ PV） power generation forecasting method based on smog factor is proposed. Multi-sensor is used to collect big data

and stepwise selection method is used to select important variables to reduce the multidimensional variables

which effectively reduces the coupling between different weather factors. The samples are clustered by hybrid Gaussian clustering algorithm. According to the result of clustering algorithm

different radial basis function neural network（ RBF） models are established

respectively. Fuzzy inference method is used to select the appropriate model. The actual prediction results validate the high precision and practicality of this method.