After the distributed photovoltaic access to the distribution network, there is a large deviation in the calculation of line loss using the traditional algorithm. Therefore, a classification and reconstruction method of a photovoltaic output curve based on daily cumulative power generation is proposed to approximately restore daily output curve. The obtained daily output curve data is used in the improved forward-backward sweep power flow method based on voltage correction to calculate the line loss of the network. Firstly, data preprocessing is carried out, and an improved spectral clustering (SC) algorithm model based on limited dynamic time warping distance (LDTW) is established. The standardized PV output curve is clustered, and the typical output modes in different seasons and weathers are obtained. Secondly, the daily cumulative power generation data collected by the power acquisition meters are classified and the output curve is linearly reconstructed. Finally, the reconstructed output curve is sampled, and the improved forward-backward sweep power flow method based on voltage correction is used to calculate the line loss of distribution network. Cases are verified based on the IEEE 33-node system, and the results are compared with the line loss results calculated by the actual photovoltaic output curve. The example shows that this method can accurately estimate line loss of distributed network by only using the daily cumulative power generation data of photovoltaic meters.