Enhancing Security Threat Management In Smart Grids Through Machine Learning-Enabled Supervised Anomaly Detection

The traditional power grid can be managed by smart grids, which are more efficient and reliable at delivering and calming energy. It introduces new openings through which adversaries might launch hacking attacks, potentially having far- reaching consequences like widespread power outages and infrastructure damage. Learning algorithms primarily depend on reinforcement approaches to detect cyberattacks in smart networks, which require frequent occurrences of attack for training and testing purposes. This study presents a unique method for detecting cyber attacks using information from power management units (PMUs) in microgrids, emphasizing semi-monitored anomaly detection and in-depth pattern identification. A Supervised Anomaly Detection in Smart Grids (SAD-SG) system trains a classifier on instances of normal circumstances to assist in detecting attack occurrences. In addition, for a supervised study's model development purposes, the training sample must reflect a variety of attacks. This research used publicly available cyber-attack systems to compare several well-known supervised outlier identification approaches. Comparing semi-monitored algorithms to well-known monitored algorithms demonstrates that the latter can find more instances of threats. Word representations also showed that the effectiveness of semicontrolled anomaly detection devices was increased by increasing their depth.