Photovoltaic (PV) power generation has the typical characteristics of intermittence and volatility. Therefore, it is of great importance to accurately predict solar output for optimization of power grid scheduling, power grid planning, and improving the competitiveness of the renewable energy power generation. Based on the modified grey back propagation (BP) neural network, this paper proposes a multi-model combination photovoltaic output power prediction method. The conventional grey model, the power function transformation grey model, the residual modification grey model and the equal-dimension-newly-information grey model are used and all single grey forecasting results are optimized combination by utilizing BP neural network. The combination weights are automatically adjusted according to the deviation of the output values and expected values. This method avoids the complex process of calculating weight coefficient. By integrating multiple single prediction results as the sample to train the BP neural network, it can finally obtain the optimal weights and accurate prediction results. The forecasting is realized based on real PV data of Hubei power grid. Example calculation demonstrates that the proposed method can significantly improve the prediction accuracy of photovoltaic output. This work is supported by National Natural Science Foundation of China (No. 51277135 and No. 50707021).