Evaluation of Anomaly Detection and Adversarial Attacks in the IEEE 9-Bus System

This study focuses on anomaly detection and the impact of adversarial attacks in microgrids, which are critical components of modern power systems. The performance of a Multilayer Perceptron (MLP) based anomaly detection model was analysed under various operating scenarios and against adversarial attacks, specifically the Fast Gradient Sign Method (FGSM) and Momentum Iterative Fast Gradient Sign Method (MI-FGSM). The performance of the anomaly detection model against adversarial attacks was examined under white-box, grey-box, and black-box scenarios with different perturbation magnitudes. It was evaluated using accuracy, precision, recall, and F1-Score metrics. The results demonstrate that white-box attacks significantly reduce the model's performance, with MI- FGSM being more effective than FGSM. Grey-box attacks had a lower impact on the model's performance compared to white-box attacks. On the other hand, black-box attacks had only a limited effect. Moreover, an increase in perturbation magnitude was found to lead to a significant reduction in the model's performance. These findings emphasize the weaknesses of MLP-based anomaly detection systems in microgrids against adversarial perturbations.