融合注意力机制和卷积神经网络的电网暂态电压稳定评估及可解释性分析

To improve the timeliness and accuracy of power grid stability assessment in complex and variable operating scenarios, a transient voltage stability assessment and interpretability analysis method integrating Attention and convolutional neural networks (CNN) is proposed. Firstly, a convolutional block attention module (CBAM) is adopted to enhance the feature capture capability of CNN, and a CBAM-CNN module is designed considering the model characteristics and network structure. Secondly, based on CBAM-CNN, the transient voltage stability assessment model is established to reveal the mapping relationship between key electrical variables and stable states of the power system under various operating conditions. Finally, based on the Shapley additive interpretations (SHAP) theory, an interpretability analysis framework for data-driven model evaluation results is established. The dominant features affecting the stable state of samples are extracted, and the influence of each input feature on the model's predictions is assessed. The accuracy of the proposed transient voltage stability assessment method and the effectiveness of the interpretability analysis method were verified in typical receiving-end power grids with voltage instability issues.