Method for detecting anomalies in acquired time series and associated devices

Abstract

Method for detecting anomalies in acquired time series and associated devices. The invention relates to a method for detecting anomalies in time series comprising: - training on normal training, normal calibration, and abnormal series, including the steps of: applying a statistical technique to the normal series to obtain training representation parameters, applying the statistical technique to the normal calibration and abnormal series to obtain calibration representation parameters, calculating, for each calibration representation parameter, a list of training distances, determining an anomaly threshold based on the consistency rate of the time series, - anomaly detection in a obtained series comprising the following steps: applying the statistical technique to the obtained series to obtain representation parameters, calculating,For each representation parameter, from a list of distances, the resulting time series is classified according to the consistency rates. Figure for the abbreviation: 1,