Abstract:With the proposal of carbon peak and carbon neutral goal in China, photovoltaic power generation plays an increasingly important role in China''s power system, and its development is related to the construction of new power system and the realization of low-carbon goals. However, due to the different installation time, equipment and operation level, the power generation efficiency of photovoltaic power stations is uneven. The horizontal comparison of the power generation efficiency of photovoltaic power stations and the diagnosis of inefficient power stations are conducive to promoting the overall development of the industry. However, due to the protection of their own data privacy, the power stations are not willing to exchange data, leading to the phenomenon of "data island". Therefore, this paper proposes to use the federated learning algorithm with privacy protection technology to evaluate and compare the power generation efficiency of different photovoltaic power stations, and deduce the average power generation efficiency, so as to further make the inefficient diagnosis of independent photovoltaic power stations. The method proposed in this paper achieves the efficiency evaluation and inefficient diagnosis on the premise of not obtaining the private data of power stations, which helps to promote the efficiency improvement and reasonable competition in the photovoltaic industry.