Compared with the traditional PV deterministic point forecast, PV interval prediction can provide more comprehensive and effective forecast information for grid dispatcher. Therefore, the paper proposes a interval prediction method based on Fuzzy information granulation theory. In view of the fluctuating character of the photovoltaic power, the original PV data is decomposed into several sub-sequences by using the ensemble empirical model decomposition（EEMD）. According to the Sample entropy theory, the sub-sequences with higher complexity are reorganized into random components, which represent the volatility of PV output. The paper conducted the random components with fuzzy information granulation, which provided its fluctuating trend, fluctuating upper bound and lower bound. And the remaining sub-sequences with relatively small complexity represent the PV stabilized components, and therefore, the deterministic predictions are made directly. In this paper, the artificial neural network model (CSO-BP), which is improved by crisscross algorithm(CSO), is used to predict the PV interval prediction results.