Development of Scan Statistics Methods for Predicting Anomalies in Information and Energy Systems

Abstract

The main objective of the research is to develop high-performance intelligent software tools for solving complex problems related to the analysis of spatio-temporal data and anomaly detection using scan statistics methods. In the field of electric power engineering, this approach has demonstrated high efficiency in forecasting non-stationary processes, such as wind power generation and price fluctuations in the wholesale electricity market (Dynamic Time Scan Forecasting method). Scan statistics methods are also actively used to detect anomalies and cyberattacks in smart grids, as well as to detect heat losses in smart heating and energy distribution systems. The article considers a classic problem, which requires finding the probability of an event consisting of the fact that when scanning an interval (0,1) with a sliding window, into which n random points are pre-dropped, the window will never contain more than a certain maximum allowable number of points. A general analytical solution to this problem remains unknown. In this paper, we proposed and implemented a comprehensive, multi-stage software approach to solving it. The advantage of the proposed approach is that, firstly, it solves an important scientific and applied problem, and secondly, the tools developed to solve it expand the theoretical foundations of scan statistics.