Abstract: In recent years, household distributed photovoltaics have developed rapidly, and the unimpressive characteristics have brought great challenges to the power sector's planning and control of the power grid and the management of user energy consumption. In view of this, this paper proposes a two-stage identification method for distributed photovoltaic output based on missing data reconstruction. First, the net load data of smart meters is clustered based on feature extraction to determine the date range of photovoltaic installation and obtain the historical load information of users. Then, a distributed photovoltaic output identification method based on missing data reconstruction technology is proposed, which transforms the distributed photovoltaic output identification problem into the actual load missing data reconstruction problem. On this basis, the photovoltaic characteristic information hidden in the net load is fully considered, the photovoltaic output identification result is corrected, and the refined identification of the target user's photovoltaic output is realized. The simulation results show that compared with other data-driven PV decomposition methods, the proposed method decreases by 1.40 kW and 0.12 kW in the standard root mean square error index, 6.80% and 1.36% in the average absolute percentage error, and 6.00% and 2.54% in the accuracy, respectively.