Traditional fault monitoring methods in electrical control systems suffer from limitations in heterogeneous data fusion, topology-correlated modeling, and dynamic adaptability. This paper proposes a multi-sensor fusion intelligent monitoring and fault localization method based on a dynamic spatio-temporal graph convolutional network (ST-GCN). In this method, the electrical system is modeled as a dynamic directed graph, the nodes fuse the characteristics of multi-source sensors such as voltage, current and temperature, and the edge weights dynamically represent the fault propagation probability. By combining the spatial attention mechanism and the time convolution module, the collaborative optimization of local anomaly detection and global propagation path analysis is realized. The experiment builds a 12-node electrical system data set based on MATLAB/Simulink platform, covering four types of typical faults and normal working conditions. The results show that the proposed method is superior to the existing methods in fault detection accuracy (94.2%), F1 score (91.8%) and root cause location accuracy (88.5%). The detection delay is as low as 150ms, and it has good robustness and interpretability, which provides a new paradigm for intelligent operation and maintenance of electrical systems under the background of Industry 4.0.
Zhu et al. (Sun,) studied this question.