Taxonomy of outlier detection methods for power system measurements

Abstract

Abstract The new emerging technologies utilize various sensors, deployed in an ad‐hoc manner to reduce energy consumption in data communication. The data collected from these sensors is huge and have a high possibility of being polluted by outliers. Therefore, researchers are trying to develop better and faster outlier detection techniques that can handle large amount of data. In this paper, the research works from the year 2000 to 2022 have been reviewed. Several fundamental and latest outlier detection methods are discussed and categorized on the basis of statistical properties, density, distance, and clustering. The other methods discussed in this paper are ensemble methods and learning‐based methods. The definitions, causes of outliers, and different methods of outlier detection are discussed. Further, one of the efficient methods from each category of the method is implemented on synthetic data of the IEEE 13‐bus distribution system. The IEEE 13‐bus system is assumed to have a Multi‐Function Meter (MFM) at each line in the system. The data captured is injected with a fixed number of outliers at a given instant. Thereafter, the performance of all the methods is tested based on the number of outliers being detected.

Keywords

• artificial intelligence applications
• big data
• information and communication
• outlier detection