A novel model called forest for photovoltaic power generation (FPPG) is proposed to effectively predict the generation power of photovoltaic power generation. FPPG is an assembly predict model composed by multi regression tree. In learning stage, FPPG first obtains training sets with diversity using random sample technique, and then, constructs regression trees with diversity. For prediction stage, the leaf that the has been assigned input information along the path of each tree predicts the power generation. Then FPPG obtains the corresponding prediction by averaging the predictions of the corresponding leaves. Experimental results show that, compared with neural network, FPPG performs better generated power forecasting.