Power system log anomaly detection based on spatiotemporal combined GAT-BiGRU

Because the existing logo detection methods are often only targeted at single features, such as quantitative relationships or sequential relationships and log module characteristics, they often ignore the characteristics and time structure characteristics of the log, which leads to a higher error reporting rate. It is proposed that the GAT-BiGRU based on a logo detection model combined with time and space makes the log abnormal detection more effective. First of all, through the improved word -level two -way door control cycle unit capture the symbol of logic and time structure combination characteristics, the structure is an orderly incremental timing event diagram, and the viewing network model (GAT) is used to extract the time to compose a log time diagram. Structural characteristics to achieve effective integration of time structural relationships and log semantic information. The TGBLog algorithm precision proposed in this paper reaches 95.67%, and the F1 value reaches 96.37%, which is improved.