Abstract: For the effective identification of faults in distributed photovoltaic（PV） systems, this study proposes a GRU-Bayesian neural network-based method for anomaly detection in the power output of distributed PV systems. Firstly, sunny days are selected as the detection scene to reduce the interference of weather factors. Then, the gray absolute correlation degree algorithm is introduced to screen out unqualified PV output data by utilizing the similarity of PV system output in the same region and construct a dataset of normal PV output for users. The GRU-Bayesian neural network is used to train and obtain the normal PV power interval for detection. Finally, actual data is used for case analysis, demonstrating the feasibility and effectiveness of the proposed method.