Passenger flow prediction is of great significance for public transportation. Most of the existing studies mainly predict the flow for a single station only extracting temporal features without considering spatial features. Passenger flow predicting for multiple stations or even the whole network is beneficial to grasp the overall situation, which has more research value in practical application. Thus, a passenger flow prediction model for multiple stations has been proposed based on spatio-temporal attention mechanism. This model is applied to Xiamen bus rapid transit (BRT) using the time granularity of 5-min. The experimental results demonstrate that our model improves the prediction accuracy compared to the baselines.