Transformer-based neural optimization for energy-efficient event detection in wide-area power monitoring systems

The security and stability of modern power systems rely heavily on timely and accurate detection of anomalies, including cyberattacks. Traditional anomaly detection approaches often struggle to capture complex interdependencies among system components, resulting in decreased detection performance. In this study, we propose a novel deep learning-based method for detecting anomalous events in power networks. The proposed model leverages historical and real-time measurement data to identify deviations that indicate potential false data injection attacks. Extensive experiments were conducted on standard IEEE test systems, incorporating adversarially injected cyberattack signals to evaluate the robustness of the approach. The results demonstrate superior performance in accuracy, precision, and recall compared with conventional autoencoder-based models. This method provides a reliable tool for enhancing the cybersecurity and operational reliability of power systems.