Intelligent Fault Diagnosis of Electrical Secondary Circuits Using Multi-Feature Fusion Deep Learning

Abstract

 To address the complex fault characteristics of electrical secondary circuits and the limited diagnostic capability of single-source information, this study proposes an intelligent fault diagnosis method based on multi-feature fusion deep learning. Current and voltage waveforms, statistical parameters, protection operations, circuit-breaker positions, and alarm states are jointly modeled. A multi-scale one-dimensional convolutional neural network extracts local transient patterns, a Transformer encoder captures long-range temporal dependencies, and an attention mechanism adaptively fuses the resulting representations. On an eight-class dataset containing 9600 samples, the proposed model achieves an Accuracy of 97.43%, an F1-score of 97.35%, and an AUC of 99.41%, outperforming representative machine learning and deep learning baselines. Ablation and noise tests further demonstrate the contributions of the feature branches and the robustness of the fusion strategy. The method provides a practical basis for intelligent maintenance and online condition monitoring of electrical secondary circuits.