Multi-dimensional KPI data anomaly detection model training method and device, and multi-dimensional KPI data anomaly detection model identification method and device

• Main Authors: TANG YULIANG, XIE HUIQIONG, WANG YILIN, YANG BO, DAI YU, HUANG LIANFEN, DING BAOGUO, WANG XIANG, YU JITAO
• Format: Patent
• Language: Chinese; English
• Published: 26.04.2024
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The embodiment of the invention provides a multi-dimensional KPI (Key Performance Indicator) data anomaly detection model training method and device and a multi-dimensional KPI data anomaly detection model recognition method and device. The method comprises the following steps: determining a correlation value between every two of a plurality of KPIs according to collected initial data of the plurality of KPIs so as to cluster the plurality of KPIs to obtain at least one KPI cluster and a central KPI thereof; the initial data of each KPI is preprocessed, and target data corresponding to each KPI is obtained; training a pre-constructed initial anomaly detection model by adopting the target data of each central KPI to obtain a target anomaly detection model corresponding to the central KPI; and according to correlation values between the central KPI and other KPIs in the same KPI cluster, migrating all or part of parameters of a central KPI anomaly detection model to obtain a target anomaly detection model corre