-
ANOMALY DETECTION IN SMART GRIDS: A MACHINE LEARNING PERSPECTIVE ON ELECTRICITY THEFT PREVENTION
Shakeel Ahmad Najar1, Dr. Ajit Kumar2
Abstract:
Data from smart grids can be examined to find anomalies in a variety of fields, including cybersecurity, fault finding, energy theft, etc. There is a compelling case to be made for anomaly detection using machine learning. We need to extract features from the raw grid data. Any occurrences or modifications in the smart grid data that deviate from the typical pattern are referred to as anomalies. The outcomes of a common grid layout can differ greatly based on patterns or modifications in power, voltage, current, or consumption. In this research, an anomaly detection model is developed for a hardware-based testbed implementation of a real-world smart grid system. It is possible to enhance system behaviour in data communication flow by identifying anomalous activity. Additionally, it will detect any changes in parameters that can point to the existence of cyberattacks. Our suggested anomaly detection methodology uses several decision trees to separate outliers from typical observations, based on the Isolation Forest (IF) paradigm. The simulation findings were used on a hardware-based testbed to validate the effectiveness of the suggested detection strategy. Principal component analysis was utilised to optimise feature selection, and the dickey-fuller test was employed to assess the model's performance.
