In order to reduce the influence of weather factors on the prediction accuracy of the output value of photovoltaic power generation, it is proposed a prediction model incorporating the clustering algorithm ( KDGMM), the improved variational modal decomposition (VMD) and the stochastic configuration network ( SCN), starting from both cluster analysis and signal decomposition. Firstly, the meteorological data are classified into sunny, cloudy and rainy days by KDGMM clustering, and for the problem that it is difficult to predict accurately on cloudy days, gray correlation analysis (GRA) is used to select similar days, and secondly, the L􀆧vy northern goshawk optimization (LNGO) algorithm is introduced to optimize VMD to get the optimal parameters, so as to reduce the nonsmoothness of PV power on cloudy days. Finally, the SCN prediction model is constructed to predict the PV power data and output its prediction results. Through experimental analysis, the root mean square error (RMSE) and the mean absolute percentage error (MAPE) of the proposed method are only 1. 44 and 1. 3%, and the R 2 index for goodness of fit is as high as 0. 99. Compared with other prediction methods, the proposed method has higher prediction accuracy