With the promotion of the distributed photovoltaic project across the country

the installed scale and scope of the distributed photo-voltaic are gradually expanding

and the reverse trend of power flow occurs in the station area. However

the existing photo-voltaic meters have no way to complete the real-time collection of power generation data

which will lead to such problems as distortion of load data at the dispatching side and the lack of distributed new energy management. Considering the measurement and communication status of primary equipment in the station area

a kind of real-time photovoltaic output estimation method based on signal decomposition theory is proposed in this paper. Based on the electric energy data of the transformer meter and the contour data of photovoltaic output in the station area

the photovoltaic output is decomposed in accordance with the minimum mutual information

and the separation results are denoised by Kalman filter. Furthermore

the database is set up based on the labeled data obtained by decomposition

and the convolutional neural network is used for training to obtain the optimized photovoltaic output estimation results. Finally

the actual data of a photovoltaic neighborhood area in Yangzhong area is verified

and the error of the algorithm is assessed.