Anomaly Location and Damage Recovery in Batch Data of Seasonal Time Series: A Case Study of User Power Data in A Province
Haijing Zhang, Hongxia Guo, Yang Yang · 2023
Time series anomaly location and damage recovery has become the focus of researchers. Yet most of the existing sequence recovery algorithms have high requirements on the source data specifications, as long as poor compatibility. In this research, an anomaly detection framework was proposed, which could be compatible with corrupted data and intact data. This task was divided into four stages, which were sequence classification, rough detection of abnormal data, precise detection of abnormal data and correction of abnormal data. Finally, the power consumption data of more than 300,000 users in a province was applied as the research case for verification. The analysis results showed that the proposed framework had high generalization and compatibility for different types of time series, and could effectively recover the source data.