The distributed photovoltaic power generation system is widely distributed and lacks a scientific and standardized operation and maintenance management system. Due to the limited availability of data, it is difficult to accurately identify abnormal conditions in photovoltaic devices caused by weather fluctua-tions.In this paper, According to the operation and maintenance status and data characteristics of distributed photovoltaic, a Quantile Regression Neural Network (QRNN) based method is proposed for detecting PV power outlier. First, the solar irradiance characteristics of sunny days are analysed, and the influence of rainy weather is excluded by using the sunny day screening method. Then, the power output correlation of different power stations is analysed to obtain the PV stations with high power output correlation as a horizontal reference, and then the power output curves of the stations to be tested on different sunny days are compared vertically to exclude the interfering factors such as weather and environmental conditions. The QRNN model feeds the meas-ured active power data to obtain the PV normal active power range, and uses the threshold to detect PV genera-tion power outliers. It can eliminate the meteorological influence, accurately identify the faulty photovoltaic system, and promote the fine operation and maintenance of distributed photovoltaic.