Solar energy is developing fast recently. Solar energy has the feature of intermittent, fluctuation and random, and it has serious harms on large scale photovoltaic grid-connected generation. This paper proposes a method to predict daily total solar-radiation and photovoltaic generation using wavelet-neural network. This method uses wavelet function to substitute the transfer function of neural-network hidden layer. In the prerequisite of not influencing forecast accuracy, this method largely shortens the practice time of model, enhances the speed of practice, and avoids neural-network getting involved in local optimal solution. Based on model of photovoltaic system, the daily total solar-radiation could be obtained binding with the prediction data of daily total solar-radiation.