%0 Journal Article
%T 基于图神经网络的电网设备拓扑感知型故障预测的算法<br>A Topology-Aware Fault Prediction Algorithm for Power Grid Equipment Based on Graph Neural Networks
%A 贾文瑞
%J Computer Science and Application
%P 128-140
%@ 2161-881X
%D 2025
%I Hans Publishing
%R 10.12677/csa.2025.156164
%X 为应对电网设备运行状态的复杂演化与多维拓扑关联特性&#65292;本文提出一种基于图卷积网络(GCN)与门控循环单元(GRU)联合建模的拓扑感知型故障预测算法。该方法以电网拓扑构建图结构&#65292;融合两类关键特征&#65306;一是设备运行状态构成的时间序列数据(如电压、电流、有功/无功功率)&#65292;二是节点的静态拓扑属性(如节点度、聚类系数与中介中心性)。GCN模块用于捕捉每一时间步的空间邻域特征&#65292;GRU模块建模节点状态的时间演化&#65292;实现对空间依赖与动态趋势的联合学习。实验结果表明&#65292;所提方法在故障预测准确率方面相较纯GCN提升4.3%&#65292;相较SVM提升12.7%&#65307;在F1分数与召回率等指标上亦明显优于对比模型。<br>To address the complex evolution and multidimensional topological dependencies of power grid equipment operation, this paper proposes a topology-aware fault prediction algorithm based on the joint modeling of Graph Convolutional Networks (GCN) and Gated Recurrent Units (GRU). The method constructs a graph structure based on the power grid topology and integrates two key types of features: (1) Time-series data representing device operational states (e.g., voltage, current, active/reactive power), and (2) Static topological attributes of each node (e.g., node degree, clustering coefficient, and betweenness centrality). The GCN module captures spatial neighborhood features at each time step, while the GRU module models the temporal evolution of node states, achieving joint learning of spatial dependencies and dynamic trends. Experimental results show that the proposed method improves fault prediction accuracy by 4.3% compared to a pure GCN model and by 12.7% compared to a Support Vector Machine (SVM). It also outperforms baseline models in terms of F1-score and recall.
%K 图神经网络&#65292
%K 故障预测&#65292
%K 电网拓扑建模&#65292
%K 时空特征融合&#65292
%K 门控循环单元<br>Graph Neural Network
%K Fault Prediction
%K Power Grid Topology Modeling
%K Spatio-Temporal Feature Fusion
%K Gated Recurrent Unit
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=117874