Time-Series Risk Assessment and Explainable Response Recommendation for Substation Auxiliary Equipment Alarms

Abstract

Alarms from substation auxiliary equipment are often distributed across multiple sources. Abnormal trends are difficult to detect in advance, and response recommendations still depend heavily on manual experience. To address these problems, this paper proposes a time-series risk assessment and explainable response recommendation method for alarm source tracing. The method constructs graph-based relations among equipment, locations, circuits, alarms, faults, and response strategies. It uses sliding windows to extract features such as threshold violation degree, trend change, and historical deviation. These features are used to calculate comparable anomaly scores. The predicted threshold violation risk is then embedded into graph nodes and used for candidate fault ranking. Starting from an abnormal equipment node or an alarm node, the system combines alarm matching, graph paths, anomaly scores, and predicted risks to generate fault explanations, risk levels, and response recommendations. Across 30 scenario-based samples covering temperature-humidity, access-control, and water leakage abnormalities, the method achieved an average early-warning lead time of 8.5 min, a localization accuracy of 83.3%, a Top-1 fault hit rate of $83.3 \%$, and a response recommendation matching rate of $86.7 \%$. The results show that the method supports trend detection, association-based explanation, and response recommendation for auxiliary equipment alarms. The method supports diagnostic analysis in substation monitoring systems.