Fault Diagnosis of Drop-Out Fuses Based on Hidden Markov Model and Graph Embedding

Traditional fault diagnosis methods relying on manual inspection and static thresholds often struggle to identify subtle or evolving faults in drop-out fuses operating within increasingly complex power distribution environments, where electromagnetic interference and dynamic operating conditions may affect monitoring reliability. To address these challenges, this paper proposes a fault diagnosis method based on a hidden Markov model and graph embedding. The proposed framework consists of three stages: constructing a power equipment graph and employing an improved graph convolutional network to extract topology-aware embedding vectors; designing a four-state hidden Markov model using multidimensional time-series data, including current, voltage, and temperature, to characterize dynamic state evolution through the Baum–Welch algorithm; and integrating graph features with HMM state sequences for precise fault identification and real-time warning. Experimental results demonstrate that the proposed method achieves a fault diagnosis accuracy of 94.21% with an F1-score of 92.74%. For non-obvious faults, the F1-score remains between 0.887 and 0.945, while the average warning lead time exceeds 17 minutes. Practical deployment further reduces the manpower required for a single fault response process to 3.6 person-hours. The proposed framework significantly enhances fault detection accuracy and response efficiency, providing an effective intelligent perception solution for power distribution systems and offering technical support for reliable monitoring and maintenance of modern electromagnetic energy infrastructures.