Abstract: Distributed photovoltaic (PV) power generation systems generally do not come equipped with a variety of sensors and monitoring devices, limiting the data available for reflecting equipment operation and conducting anomaly detection. This article proposes a PV anomaly detection method based on the STL-Bayesian spatio-temporal model, which utilizes the spatio-temporal transferability of meteorological data to uncover the correlation of PV power output and perform anomaly detection. Firstly, the seasonal and trend decomposition using Loess (STL) is employed to decompose the PV active power time series data into three components. Then, the influence of different lengths of input data on the decomposition results and the spatio-temporal distribution characteristics of the components within the region are investigated. Subsequently, Bayesian models are constructed to perform short-term and ultra-short-term spatial interpolation on the trend component and the residual component, respectively, so as to obtain the PV output within the region. Finally, the earth move's distance (EMD) between the actual values and regression values is calculated to detect abnormal states. The analysis of the algorithm shows that the model has a high accuracy in the detection of both reversible and irreversible anomalies under distributed PV scenarios.