The modeling and prediction of photovoltaic(PV)power curve are a widely concerned issue in the power grid operation and planning. Mining and utilization of the big data recorded by the PV stations provide a new and promising direction for the modeling of the fluctuating PV power. A multi-scale analysis scheme is presented to realize the digging and modeling for the fixed and stochastic characteristics of the PV power time sequence. By combining the multiple cluster analysis at large time-scale

the standardized transformations at small time-scale

and the probability distribution modeling

the digging to the fluctuated and regular characteristics of PV power is realized. The clustering results are also verified by cross validation with the theoretical models in the meteorology field. The effectiveness of the proposed method and model is verified by an example of day-ahead PV power forecasting.