| 摘要 |
PROBLEM TO BE SOLVED: To solve the problem that, in abnormality detection on the basis of a multi-dimensional time series sensor signal, in spite of need to specify which sensor is related to abnormality for determining following actions such as countermeasure and investigation, there are cases where errors occurs in a conventional method to specify a sensor having large difference between a reference value and an observed value as an association sensor.SOLUTION: An abnormality diagnostic method comprises the steps of: extracting a multidimensional feature vector from a sensor signal at every time; calculating a reference feature vector of each time on the basis of a collection of a feature vector during a determined learning period and a feature vector of each time; calculating anomaly measure on the basis of difference between the feature vector of each time and the reference feature vector; detecting abnormality by comparison between the anomaly measure and a predetermined threshold; and specifying a sensor related to the abnormality on the basis of two dimensional distribution density of feature values about time when the abnormality is detected. |
