A Spatiotemporal Feature Fusion Based Method for Detecting Voltage Anomalies in Insulator Strings

Insulators, which are crucial components in high-voltage transmission lines, can degrade over time due to exposure to various environmental factors and mechanical stresses. This degradation can lead to malfunctions in the power grid, so it’s important to be able to detect abnormalities in insulator operation. Based on the idea of reconstruction, this paper proposes a novel and accurate anomaly detection model , which combines Transformer and VAE, for detecting insulators in abnormal states using multidimensional voltage time series data collected from insulator voltage sensors. The model utilizes a Transformer as a feedforward network for VAE to effectively extract and combine feature information from both the feature variables and time series, improving the ability to distinguish abnormal data from normal data through better reconstruction. Experimental results demonstrate that the Transformer-VAE(T-VAE) model outperforms three other reconstruction-based anomaly detection models in terms of precision, recall, and F1-score.