طراحي استراتژي دفاعي در برابر حملات سايبري و فيزيكي سايبري در سيستم قدرت

Defense strategy against cyber an‎d cyber-physical attacks in power system

مهندسي برق

Nowadays, the widespread integration of communication infrastructures has transformed modern power systems into cyber–physical networks, making them increasingly vulnerable to cyber threats —particularly False Data Injection (FDI) attacks—a critical security concern. These attacks manipulate measurement data, disrupt protection an‎d control operations, an‎d remain difficult to detect due to their ability to remain stealthy against conventional bad data detection mechanisms. In this dissertation, a realistic model for FDI attacks is developed in which the attacker has access only to a local region, an‎d the attack is confined to that area. The attack design is based on DC state estimation combined with voltage sensitivity analysis with respect to active power, eliminating the need for complex nonlinear AC models. The attack is initially formulated as a bi-level optimization problem. At the upper level, the attacker aims to overload a target transmission line with minimal data manipulation while remaining hidden from the operator, whereas at the lower level, the operator attempts mitigates the attack impact by solving an economic dispatch problem. To defend against such attacks, a deep learning framework based on a multi-scale transformer architecture is developed to simultaneously capture temporal, spatial, an‎d regional dependencies in system measurements. Considering the defender’s design, the original bi-level attacker–operator model is extended to a tri-level framework. The proposed model jointly performs four key tasks: attack detection, identification of the compromised region, localization of affected buses, an‎d recovery of corrupted measurements. The recovery module employs a hierarchical gating mechanism with dynamic weighting through a multi-task loss function to enhance robustness. The performance of the proposed framework is eva‎luated on the IEEE 39-bus power system, an‎d simulation results demonstrate that the proposed method achieves higher detection accuracy an‎d superior data recovery performance compared to conventional machine learning an‎d deep learning approaches.

6

166435