Accurate power load forecasting is crucial to the safe and economic operation of modern power systems. Power load forecasting can be expressed as a multivariable time series forecasting problem with certain potential spatial dependence. However， most existing power load forecasting work fails to explore this spatial dependency relationship. Based on this， this paper proposes a short-term power load forecasting method based on the spatiotemporal graph attention network. A spatiotemporal graph-based attention network module is proposed， which uses a graph attention layer to adaptively capture potential spatial dependencies between users. At the same time， a gated convolutional attention layer is used to adaptively fit the electricity consumption of each user in the time dimension to improve the prediction accuracy of the network. Actual data experiments show that the overall prediction accuracy of the model proposed is significantly improved， especially in alleviating the problem of deteriorating long-range prediction accuracy to a certain extent， verifying the effectiveness and feasibility of the proposed method.