The photovoltaic power generation depends on the uncertain external environment, therefore, the generated power has a relatively high fluctuation resulting a low ratio power access. The accurate forecasting of photovoltaic power can raise up its penetration and also keep a safe operation and dispatching of the power grid.In this paper, a prediction method for photovoltaic power is proposed based on information fusion theory. The dynamic neural network model is adopted. Then, the influence factors, which limit the photovoltaic power generation, are fully considered and a comprehensive influence factor lambda is proposed. After that, the gradient algorithm is utilized for optimization. For the period with high fluctuation, the forecast time interval is shorted and the number of layers is increased to improve the prediction accuracy. Finally, the feasible and effective of the proposed theory is verified at Yueyang photovoltaic power plant in practice.