Investigation of the Influence of the Preliminary Digital Filtering Method on the Accuracy of Signal Prediction in Anomaly Detection Systems in Industrial Automatic Control Systems (IACS)

Abstract

The technology for predicting dynamic data flows plays an important role in building systems for detecting anomalies in protecting the information of various industrial automatic control systems (IACS). In order to improve the quality of forecasting dynamic data flows, it is proposed to apply preliminary digital processing (filtering) of signals to decompose the observed time series coming from the sensors of the IACS into separate components, that is, to conduct a preliminary structural analysis of the observed time series. With this approach, component decomposition of the initial signal and serial noise filtering using a parallel set of digital filters are performed, which (as shown in the investigation) significantly improves the quality of the generated forecast. The study also conducted an analysis of the error of the forecast result, that is, the signal of the difference between the original signal and its forecast, using digital spectral and bispectral analysis. It is assumed that for the case of “perfect prediction” the prediction error is an unpredictable residue, that is, tends to a state of white noise. The paper shows that the analysis of forecast errors using digital spectral and bispectral analysis methods allows you to form an assessment of the quality of the forecast result. In the study, a comparison of the results for the case of generating a forecast using and without using preliminary digital filtering was carried out on the same initial test signal. The structure of the compared neural networks (generating the forecast) was the same and the training data sets and the target data sets for generating the forecast were identical. The comparison shows a significant increase in the efficiency of using preliminary digital filtering in order to increase the accuracy of forecasting. Work with neural networks was carried out in the MATLAB “Deep Learning Toolbox” expansion pack. For spectral and bispectral analysis, the “Higher Order Spectral Analysis Toolbox” package was used.