Визначення аномалій у часових рядах на основі візуалізації рекурентних діаграм

Abstract

Recurrence is a fundamental property of dynamical systems, which can be exploited to characterize the system’s behavior in phase space. A powerful tool for their visualization and analysis is the recurrence plot. Methods basing on recurrence plots have been proven to be very successful especially in analysing short, noisy and nonstationary data, as they are typical. This article describes a method for finding anomalies in a time series based on dividing a time series into parts. Time series parts are converted to recurrence plots - a black and white image. Further, the convolutional neural network is used to classify the image. The results showed that the considered method has a high enough classification accuracy.