Unmasking Stealthy Attacks on Nonlinear DAE Models of Power Grids

TL;DR

The paper investigates stealthy false data injection attacks on power grids modeled by nonlinear differential algebraic equations (NDAEs), showing that coupling between dynamic generator states and algebraic power-flow constraints constrains attacker influence and challenges conventional detectors. It introduces two attack classes—SCUAs (constraint-unaware) and SCAAs (constraint-aware)—and develops the Iterative Constraint-Aware Attack Algorithm (ICAA) to refine attack vectors while satisfying physical constraints. Through simulations on the IEEE 39-bus system, the work demonstrates that NDAE-based state estimation and constraint validation significantly reduce attack impact compared to simpler models, and that dynamic detectors like CUSUM outperform static ones like chi-squared under NDAE-aware conditions. The findings underscore the value of NDAE-level modeling for accurate security analysis and motivate NDAE-specific detection and mitigation strategies for resilient smart grids.

Abstract

Smart grids are inherently susceptible to various types of malicious cyberattacks that have all been documented in the recent literature. Traditional cybersecurity research on power systems often utilizes simplified models that fail to capture the interactions between dynamic and steady-state behaviors, potentially underestimating the impact of cyber threats. This paper presents the first attempt to design and assess stealthy false data injection attacks (FDIAs) against nonlinear differential algebraic equation (NDAE) models of power networks. NDAE models, favored in industry for their ability to accurately capture both dynamic and steady-state behaviors, provide a more accurate representation of power system behavior by coupling dynamic and algebraic states. We propose novel FDIA strategies that simultaneously evade both dynamic and static intrusion detection systems while respecting the algebraic power flow and operational constraints inherent in NDAE models. We demonstrate how the coupling between dynamic and algebraic states in NDAE models significantly restricts the attacker's ability to manipulate state estimates while maintaining stealthiness. This highlights the importance of using more comprehensive power system models in cybersecurity analysis and reveals potential vulnerabilities that may be overlooked in simplified representations. The proposed attack strategies are validated through simulations on the IEEE 39-bus system.

Figures (8)

• Figure 1: Single-line diagram of the IEEE 39 Bus Network.
• Figure 2: Comparison of power system states of generator 2 under SCUAs (left column) and SCAAs (right column). From top to bottom: (1) SE error norm, (2) Frequency, (3) Bus voltage magnitudes, (4) Generator rotor angle and (5) Generator active power outputs. Solid black lines represent true states, while red dashed lines show estimated states under attack.
• Figure 3: Comparison of the Mean Absolute Error (MAE) and Absolute Error Dynamics for CUSUM SCUA and SCAA/ICAA.
• Figure 4: Algebraic constraint violations over time during 50 SCUA attempts using different targetted buses.
• Figure 5: Comparison between SCAA optimization and SCAA-ICAA in 100 differnet runs under chi-squared detection settings