Implication of Unobservable State-and-topology Cyber-physical Attacks

Jiazi Zhang, Lalitha SankararXiv 2015 · OpenReview

TL;DR

This paper introduces a novel type of cyber-physical attack that can simultaneously alter the state and topology data of a power grid's control system. By cleverly masking a physical line trip with a cyber attack, attackers can force specific lines to overload, potentially leading to cascading failures, even with limited resources and detection capabilities.

What Problem Does This Paper Address?

The electric power system is a complex network where a cyber layer (SCADA/EMS) monitors and controls the physical layer. While physical attacks on power lines are a known threat, and cyber attacks like False Data Injection (FDI) have been studied, there's a growing concern about coordinated cyber-physical attacks. These attacks could combine physical sabotage with sophisticated cyber deception to achieve more severe consequences. Specifically, this paper addresses the challenge of unobservable state-and-topology cyber-physical attacks. These attacks aim to not only cause physical damage (like tripping a line) but also to manipulate the system's perceived state and topology in the control center, making detection difficult. The authors focus on understanding the physical consequences of such attacks, particularly the potential for line overloads, which can lead to cascading failures.

Methods

The core of the paper's contribution lies in its novel attack strategy, formulated as a two-stage optimization problem.

  • Attack Formulation: The attack involves two main phases:
  • Physical Attack: An attacker physically trips a transmission line.
  • Cyber Attack: Simultaneously, the attacker launches a cyber attack to:
  • Alter the line's status in the SCADA data to make it appear operational.
  • Modify measurements within a sub-network to mask the physical attack and create a false system state and topology.
  • The ultimate goal is to cause an overload on a different target line.
  • Two-Stage Optimization: To determine the optimal attack strategy, the authors employ a two-stage optimization approach:
  • Stage 1: This stage aims to identify the "switching attack line" (the one to be physically tripped) and an initial attack vector (c). It maximizes the power flow on a chosen target line while respecting attacker constraints like the number of lines that can be tripped (NT), the number of buses whose states can be manipulated (No), and the allowable load shift (TPD). This stage also incorporates a DC Optimal Power Flow (DC OPF) to minimize operational costs and ensure system balance.
  • Stage 2: This stage determines the initial attack vector (c^0) for the first cyber attack event (A1). It ensures that the resulting load shifts lead to the optimal generation dispatch determined in Stage 1.
  • Unobservability: The attacks are designed to be unobservable by the system's state estimation (SE) and bad data detection mechanisms. This is achieved by carefully crafting the attack vectors to manipulate measurements within a specific sub-graph (S) of the network, ensuring that the changes are consistent with the system's expected behavior or masked by the false topology.
  • AC vs. DC Models: While the optimization problems are initially formulated using DC approximations for tractability, the paper outlines a procedure to construct the corresponding AC attacks, which are more realistic for power system analysis.

Datasets

The authors evaluate their attack strategy on the IEEE 24-bus reliable test system (RTS). This is a standard benchmark system for power system analysis. They assume the system is operating under optimal power flow and that the attacker has knowledge of historical load patterns and generation dispatch.

Key Insights and Ideas

  • Synergy of Physical and Cyber Attacks: The most significant insight is how combining physical and cyber attacks can create a far more potent threat than either alone. By masking physical actions with cyber deception, attackers can achieve goals (like line overloads) that would be difficult or impossible with isolated attacks.
  • Unobservability is Achievable: The paper demonstrates that it's possible to design attacks that are unobservable by standard state estimation and bad data detection techniques, even when manipulating both system states and topology.
  • Targeted Overloads: The attack strategy is specifically designed to cause overloads on chosen lines, which is a critical precursor to cascading failures in power grids.
  • Limited Resources, Significant Impact: The attack strategy is effective even with constraints on the attacker's resources, such as the size of the sub-network they can access and the number of lines they can physically trip.
  • Vulnerability of AC State Estimation: The research highlights that AC state estimation, while more realistic, can still be vulnerable to these sophisticated, coordinated attacks.

Key Results

  • High Success Rate: The proposed attack strategy successfully caused line overloads in 86.84% of the tested target lines in the IEEE 24-bus system using the nonlinear system model.
  • Efficient Sub-graph Usage: For 72.73% of successful attacks, the required sub-graph for the attack had fewer than 16 total buses, indicating that attackers don't need extensive access to the entire grid.
  • State-Preserving Attacks: In some cases (lines 9 and 11), the attacks could be launched with only state-preserving topology changes, as described in prior work.
  • Tracking Linear to Nonlinear: The load shifts observed in the nonlinear system model closely tracked those in the linear DC model for most successful attacks, validating the effectiveness of the optimization approach.
  • Cascading Failure Potential: The paper illustrates how sustained overloads on a target line can lead to it overheating and tripping offline, a critical step in initiating cascading failures.

Limitations and Future Work

  • Attacker Knowledge: The current model assumes the attacker has knowledge of the entire network's topology and generation costs, and historical load patterns. Future work could explore scenarios where the attacker has more limited information about the system.
  • Sub-network Access: While the attack is confined to a sub-network for data manipulation, the attacker still needs knowledge of the global topology and generation data. Investigating attacks with more restricted global knowledge is a future direction.
  • Multiple Attacks: The paper focuses on a single target line and a limited number of physical attacks. Future research could explore worst-case attacks that trip multiple switching lines and maximize power flow on multiple target lines simultaneously.
  • Countermeasures: While the paper suggests forecasting expected generation dispatch as a potential countermeasure, further research into robust detection and mitigation strategies for these specific types of cyber-physical attacks is needed.

Summary updated

Abstract

This paper studies the physical consequences of a class of unobservable state-and-topology cyber-physical attacks in which both state and topology data for a sub-network of the network are changed by an attacker to mask a physical attack. The problem is formulated as a two-stage optimization problem which aims to cause overload in a line of the network with limited attack resources. It is shown that unobservable state-and-topology cyber-physical attacks as studied in this paper can make the system operation more vulnerable to line outages and failures.
BibTeX
@misc{zhang2015implication,
  author = {Jiazi Zhang and Lalitha Sankar},
  title = {{Implication of Unobservable State-and-topology Cyber-physical Attacks}},
  year = {2015},
  eprint = {1509.00520},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1509.00520},
}
Figure