Research on AI-based fault diagnosis methods for smart terminals in distribution networks

Smart terminals in distribution networks operate long-term within complex electrical and communication environments, making them susceptible to factors such as sampling link drift, instrument transformer saturation, protection logic disorder, and communication anomalies. Their fault characteristics exhibit multimodality, strong temporal dependencies, and cross-terminal correlations, rendering traditional diagnostic methods prone to unstable identification under complex disturbances. This study proposes a fault diagnosis method for smart terminals. It employs a multi-source data fusion strategy to unify representations of voltage/current waveforms, event sequences, and communication metrics. A temporal attention mechanism extracts abrupt changes and weak structural anomalies, while graph convolutional networks model inter-terminal correlation patterns. Experimental results based on real multi-category operational data demonstrate that this method achieves high recall and low false detection rates across six typical fault scenarios. Compared to traditional models, it exhibits enhanced robustness, generalization capability, and sensitivity to complex fault chains. The research findings hold significant implications for improving the operational reliability of distribution automation terminals and provide an extensible technical pathway for online health monitoring and adaptive control of intelligent distribution equipment