Joint Adversarial Example and False Data Injection Attacks for State Estimation in Power Systems.

Area of Science:

• Electrical Engineering
• Cybersecurity
• Artificial Intelligence

Background:

• State estimation is crucial for power systems but vulnerable to false data injection attacks (FDIAs).
• Deep learning models used for FDIA detection are susceptible to adversarial attacks.
• Existing methods struggle to bypass both traditional and AI-based defenses.

Purpose of the Study:

• To develop advanced adversarial false data injection attacks (AFDIAs) for power system state estimation.
• To explore attack scenarios that are stealthy to both bad data detectors (BDDs) and deep learning detectors.
• To optimize the impact of these joint attack methods.

Main Methods:

• Proposed a state-perturbation-based AFDIA (S-AFDIA) to ensure stealthiness against BDDs.
• Developed a measurement-perturbation-based adversarial FDIA (M-AFDIA) effective against deep learning detectors.
• Generated malicious data stealthy to both BDDs and deep learning-based detectors.

Main Results:

• S-AFDIA successfully performs attacks stealthy to both conventional BDDs and deep learning detectors.
• M-AFDIA succeeds when only deep learning-based detectors are employed.
• Proposed methods demonstrate superior performance compared to state-of-the-art techniques.
• Attack effectiveness can be optimized through joint attack strategies.

Conclusions:

• The developed S-AFDIA and M-AFDIA methods offer robust solutions for creating sophisticated FDIAs.
• These findings highlight critical vulnerabilities in current power system security frameworks.
• The research provides valuable insights for developing more resilient power grid defenses against cyber threats.