Photovoltaic (PV) power has the characteristics of large fluctuation, and its accurate prediction is of great significance for largescale PV power and grid connection. The correlation analysis and time series moving average (MA) methods are used to determine and predict weather data in the region where a power station is located, and a more accurate prediction value of the atmospheric information of PV power generation site is obtained. Principal component analysis is used to reduce dimension of meteorology data, and several key influencing factors are obtained. Finally, the improved support vector machine (SVM) algorithm is used to build the model of the relationship between multivariable feature sequence and PV power. In the verification experiment, the trained SVM model is used to complete the prediction, and the generation of prediction error is analyzed. The prediction effects between the neural network algorithm and the others are compared. The results show that the error of MASVM method is relatively small, which proves the validity of prediction.