Assessment of the State of Complex Technical Objects Based on the Analysis of Time Series Anomalies

The problem of estimating the state of complex technical objects based on the analysis of time series anomalies is considered. It is shown that to solve this problem it is necessary to use time series of large dimensionality, which have a number of features. Such features include: noise accumulation, false correlation and random endogeneity. In this regard, the analysis of experimental series obtained as a result of continuous long-term observations requires the application of special methods of artificial intelligence, which can improve the results of decision-making in poorly structured situations through machine learning. One of such methods, which has recently become widespread, is time series anomaly analysis. The study of methods for analyzing time series anomalies allowed us to establish that the most adapted for working with contextual anomalies is a group of spectral methods. We propose a method for analyzing time series anomalies based on the decomposition of the time characteristics of the studied objects into a spectral series by a special basis of exponential functions.