Missing-Tolerant Anomaly Detection for Gateway Electrical Energy Metering Device Based on Improved Transformer

The accurate measurement of the gateway electrical energy metering device is significant for the fairness of the electric energy trade settlement, and its anomaly detection has become a focus in the smart grid. Complex environments and data transmission errors often cause missing monitoring data, which causes an inconsistent data sequence length. To solve this problem, a missing-tolerant anomaly detection algorithm based on the transformer is proposed, providing a mask mechanism to cover missing data. To enhance the weakened data feature caused by missing, a topological feature encoding strategy (TFES) and a parallel perception strategy of spatiotemporal features (PPSSF) are proposed to characterize the complex topology and correlation of monitoring data. In addition, an autoregressive strategy is used to capture local data features accurately. The experimental results show that the three improved strategies effectively enhance the algorithm performance during missing-tolerant processing, and the algorithm can meet industrial application requirements when the data missing rate is within 5%. The missing-tolerant anomaly detection algorithm helps for the device maintenance and repair applications.