Abstract: Accurate photovoltaic power prediction is of great significance to the stable operation of power systems. Aiming at the problems of long operation time and poor feature extraction ability in existing forecasting algorithms when dealing with multi-dimensional input weather variables, this paper proposes a forecasting algorithm based on multi-kernel function support vector machine with parameter optimization. First of all, the new algorithm preprocesses the data, gray correlation degree is used to extract historical days with high similarity to the forecasted day to improve the forecasting accuracy, and principal component analysis（PCA） reduces the dimensionality of the input data to improve the accuracy of photovoltaic power forecasting. Secondly, in view of the relatively poor ability of single-kernel support vector machine to extract multi-dimensional data features, a multi-kernel support vector machine model is established based on linear kernel function and radial basis kernel function to predict photovoltaic power generation. Different weights are calculated according to the prediction error of each kernel function support vector machine to enhance the feature extraction ability of the input data and improve the prediction accuracy. The gray wolf optimization（GWO） algorithm is used to determine the parameters of the support vector machine with different kernel functions to improve the prediction accuracy. Finally, the prediction effect of the algorithm is verified by the historical dataset of a photovoltaic power station in Beijing. The example analysis shows that compared with the traditional forecasting algorithm, the forecasting accuracy and speed are significantly improved.