Abstract: Photovoltaic (PV) output power has randomness and uncertainty, which makes it difficult to establish an accurate prediction method. However, compared with the traditional deterministic point prediction method, PV output range prediction is more important for the safe and stable operation and economic dispatching of power system. Therefore, we propose a day-ahead interval prediction model of PV power generation output based on the complementary ensemble empirical mode decomposition (CEEMD) and deep belief network (DBN) optimized by simulated annealing (SA). Firstly, the similar daily sample sets of sunny and cloudy weather are selected through the similarity coefficient of comprehensive factors. On this basis, CEEMD is used to decompose the PV output sequence into several components with different features. Then SA-DBN and kernel density estimation (KDE) are used to predict the PV output power in the day-ahead interval. Finally, the effectiveness of the proposed method is verified by the example data. The results of historical data analysis of several PV power stations show that the proposed model can give the confidence interval based on the error distribution more accurately and is not affected by the geographical location of PV power stations.