(Translated by kipris) The present invention provides an abnormal value detection device of a cyber-physical system which comprises: a learning module that learns a self-organization map by learning multivariate time series data; a calculation module that calculates a distance function between a codebook matrix output of the self-organization map and real-time data and replaces the distance function with a Frobenius norm to calculate statistics when inputting the real-time data; and a detection module that detects an abnormal value of the real-time data by performing real-time multiple testing on the statistics.COPYRIGHT KIPO 2024