Due to the volatility of photovoltaic power generation,the existing photovoltaic power generation prediction technology has problems such as incomplete consideration of meteorological factors and insufficient feature extraction. In order to improve the accuracy of photovoltaic power generation prediction,an improved typical meteorological year method (TMY Method) is proposed to generate typical meteorological year data,and this method is combined with the generalized regression neural network (GRNN) to predict photovoltaic power generation. First of all,six kinds of historical meteorological indicators are selected,and Finkelstein-Schafer statistical method is used to select typical meteorological week and generate typical meteorological year data. Then,the factor analysis method is used to filter out the meteorological indicators that affect the photovoltaic power generation,and the selected meteorological indicators and daily photovoltaic power generation are standardized as the initial input of the GRNN model to obtain the predicted daily photovoltaic power generation. Finally,the designed model is trained and predicted by historical weather data and daily power generation data from Nanjing,Jiangsu Province. The results show that the prediction method proposed in this paper has better prediction accuracy than the original TMY-GRNN prediction method dose.