Abstract: The gradual increase of photovoltaic installed capacity has brought great challenges to the large-scale photovoltaic gridconnection, and the photovoltaic power prediction beyond the longer forecasting period is conducive to the safe and stable operation of the power system. The longest forecasting period of existing research and application is 7 days, in order to extend the forecast period to 8～15 days, a power forecasting method for photovoltaic cluster in long forecasting period is proposed based on the power reconstruction and temporal characteristic constraints. Firstly, the daily electricity quantity and irradiation energy are calculated by approximate integral. Secondly, the variational mode decomposition（VMD） is optimized based on sparrow search algorithm（SSA） to decompose the electricity quantity and irradiation energy sequence. Multiple linear regression model is used to forecast and superimpose the components of different frequencies to obtain the electricity quantity forecasting results. Then, a constraint process is established according to the output characteristics, and the electricity quantity forecasting results are reconstructed into photovoltaic power. Finally, the proposed method is applied to a photovoltaic cluster in Gansu Province, China, and the root mean square error of the power forecasting for the proposed model is reduced by 2.55% on average in typical months in different seasons, which verifies the effectiveness of the proposed method.