According to the output power of PV generation system having the characteristics of non-stationary and randomness, a forecasting model for grid-connected photovoltaic generation system output power is proposed based on EMD and SVM optimized by ABC algorithm. Firstly, the time series data of output power in the similar day with the interval of 15 minutes is built on the basis of weather forecast data of the forecast day. Then, the time series data of output power is decomposed into a series of components including some intrinsic mode components and a trend component under different scales by using EMD, and different SVM forecasting models are built for each intrinsic mode components and trend component, and the parameters of SVM model are optimized by ABC algorithm. Finally, the entire forecasting results are combined into the ultimate forecasting result of grid-connected photovoltaic generation system output power. The forecasting model is tested with the field dada and the results show that the model based on EMD-ABC-SVM has higher accuracy and faster speed compared to single SVM model and EMD-SVM without optimization.