Anomaly detecting method, and apparatus for the same

Abstract

In an anomaly detecting method by using multi-dimensional time series sensor signals including, generating anomaly model by using data of a learning period including neither that period nor any exclusion candidate period, calculating anomaly measurements on the basis of the distance from the normal model and, making a period containing the maximum anomaly measurement value but no exclusion candidate period, learning exclusion periods and anomaly determining thresholds are determined as learned data on the basis of the result in each round, generates anomaly model data in a learning period except learning-exclusion periods regarding acquired data or data in a designated evaluation period, an anomaly measurement at each time point is calculated on the basis of the distance from the normal model, and data at each time point is determined to be anomaly or normal by comparing the anomaly measurements with anomaly determining thresholds.