Secure AGC Systems: Attack Detection via a Novel OC-SVM Ensemble Learning Framework

The explore aims to provide a Semi-supervised Learning method for spotting irregularities in power systems' Automatic Generation Control loops. An ensemble approach using a Gaussian One-class SVM, K-Means, and a novel technique known as REO is proposed as the methodology. The proposed method differs from previous work in this field as it can learn from just vigorous class information that is voluntarily obtainable in ancient databases or may be gained via offline reproduction. During training, the attack samples are not used since it is advised that the algorithm be built with the notion that defenders can't predict the assault sequences in advance. To see how well the suggested method identifies attacks that haven't been seen before, we've produced two additional kinds of intelligently generated attacks on top of the usual False Data Injection attacks. Initially, a well-coordinated stealth attack is planned, which might compromise the stability of the system. Coordination and secrecy characterise this assault. Second, a new strategy for financial gain-oriented market operation attack is created. More cash settlements for the attacker-selected producing firms are the goal of this assault. Lastly, we compare the recommended ensemble method's performance to that of existing one-class classifiers. The comparison shows that the suggested ensemble framework outperforms the alternatives with higher TPR (99.8%), TNR (98.9%), and Accuracy (99.4%).