Accurate and reliable distributed photovoltaic (PV) power prediction can play a key supporting role for active distribution network operation management, dispatch operation, and distributed new energy consumption. Considering the strong correlation of neighboring PV power data at the same moment, this paper proposes a distributed PV power prediction method based on station correlation and data-driven. Firstly, the dynamic Self-Organizing Map algorithm clusters the data of each distributed PV field station to obtain a cluster data set of PV field stations with similar power output characteristics. On this basis, the Gated Recurrent Unit network-Bayesian combination model is used to dig deeper into the PV output characteristics among the associated stations, and the PV power interval prediction is carried out using the Bayesian optimization parameter tuning. Finally, based on the real distributed PV power data in a certain area, an example analysis is conducted to confirm the efficacy of the approach in this paper.