Considering the randomness and volatility of photovoltaic output power, a short-term power generation output prediction method is proposed based on optimizing a Deep Belief Network (DBN) with Particle Swarm Optimization (PSO). First, the PSO algorithm is used to determine the optimal initial weight of the DBN neural network, and the initial DBN network is established. Second, after the prediction date is determined, the grey correlation degree method is used to select the date with high similarity to the predicted daily meteorological features. Meteorological data and historical power generation data are used as training sets to train the initial DBN network and establish a prediction model. Finally, the simulation results show that the model used in this paper has higher prediction accuracy than the traditional DBN neural network. This work is supported by National Natural Science Foundation of China (No. 51477070).