Abstract: To accurately identify critical nodes to ensure the reliable operation of the power system, a method is proposed in power grids using a dual self-attention convolutional neural network (CNN) that integrates topological features and electrical features. First, a multi-dimensional feature set was constructed, including local topological features, semi-local topological features, electrical distance, and node voltage. Second, the CNN was improved using a squeeze-and-excitation (SE) self-attention mechanism to enhance node feature extraction capabilities, and a Transformer encoder with multi-head self-attention was introduced to achieve deep fusion of topological features and electrical features. The results on the IEEE 30-bus and IEEE 118-bus standard test systems demonstrate that the proposed method identifies critical nodes with higher accuracy. In terms of node influence assessment and network robustness, the critical nodes identified by this method have a stronger impact on the network and enhance its robustness, providing effective decision support for the secure and stable operation of the power grids.