Analysis of Point and Collective Anomalies in Energy Consumption Data

The article describes various methods for detecting point and collective anomalies in energy consumption data. The authors suggest a construction method for the typical daily power consumption profile, any deviations from which are considered anomalies. In this paper, to detect point anomalies, well-known algorithms for determining outliers in data and the DBSCAN clustering method were used. These algorithms produce similar results and can be used independently or combined to analyze point anomalies. The article also describes a method that allows detecting collective anomalies in energy consumption data. To identify collective anomalies, some data generated according to specified templates were added to the data sets. After that, the initial data set was recoded as a collection of monochrome images, marked up, and recognized using a two-dimension convolutional neural network. While solving a multi-class classification problem, the suggested method helps to detect anomalies and their types and, therefore, can identify their potential causes.