Modeling Load Redistribution Attacks in Integrated Electricity-Gas Systems

Rong-Peng Liu, Xiaozhe Wang, Bo Zeng, Rawad Zgheib

TL;DR

This paper introduces a sophisticated bilevel mathematical model to identify the most damaging "load redistribution" (LR) cyberattacks on integrated electricity-gas systems (IEGSs). By incorporating unit commitment and a novel decomposition algorithm, the researchers demonstrate how stealthy attackers can exploit system congestion to cause significant economic losses and load shedding.

What Problem Does This Paper Address?

As power grids increasingly rely on gas-fired generation, the electricity and natural gas sectors have become deeply interdependent. This "Integrated Electricity-Gas System" (IEGS) relies heavily on cyber infrastructure for coordinated operation, creating a broad surface for cyberattacks.

The authors focus on Load Redistribution (LR) attacks. Unlike crude attacks that simply shut down components, an LR attack is a type of False Data Injection (FDI) where an intruder manipulates load meter readings. By making the load appear higher in one area and lower in another—while keeping the total load count the same—the attacker can trick the system operator into making sub-optimal dispatch decisions. These attacks are notoriously difficult to detect because they don't violate basic power-balance checks.

Previous research mostly focused on power-only systems. Modeling these attacks on IEGSs is much harder due to the non-convex physics of gas flow (the Weymouth equation) and the risk that a mathematical model of an attack might become "infeasible" (i.e., the math breaks) if it doesn't account for how real-world operators respond to stress.

Methods

The authors propose a Bilevel Mixed-Integer Linear Program (MILP):

  • Upper Level (Attacker): Seeks to maximize total system cost (damage) by injecting false data into power and gas load measurements, subject to "stealthiness" constraints.
  • Lower Level (Operator): Seeks to minimize system costs (Economic Dispatch) based on the falsified data.

Key Innovations: 1. Augmented Model with Unit Commitment (UC): Standard models often assume all generators stay on. The authors prove that by including binary UC variables (allowing generators to shut down), the model remains mathematically feasible even under extreme attacks. This ensures the attacker's "worst-case" scenario is always found. 2. Weymouth Linearization: The non-convex gas flow equations are handled using an incremental Piecewise Linear (PWL) approximation, making the problem solvable with standard MILP solvers. 3. Modified Reformulation and Decomposition (R&D) Algorithm: Bilevel MILPs are notoriously difficult to solve. The authors developed a master-subproblem framework that uses KKT (Karush-Kuhn-Tucker) conditions and specific "cuts" to ensure the algorithm converges to a global optimum even when traditional assumptions about bilevel problems fail.

Datasets

The study evaluates the model on two benchmark systems:

  • IEGS-6-7: A small-scale system with a 6-node electricity grid and a 7-node gas network.
  • IEGS-39-20: A medium-scale system consisting of a 39-node electricity grid and a 20-node gas network.
  • The system data and topologies were sourced from established literature (Reference [38] in the paper).

Key Insights and Ideas

  • The "Infeasibility" Insight: The authors discovered that if a model doesn't allow for unit commitment or load shedding in the operator's response, the math might suggest an attack is "impossible" simply because the grid would crash. In reality, an operator would shed load or shut down plants, which is exactly what the attacker wants. Including these options makes the attack model more realistic and dangerous.
  • Congestion-Related Cyberattacks: The results show that LR attacks are most effective when the system is already operating near its transmission or pipeline limits. If a system has "infinite" capacity, redistributing load has no economic impact.
  • Gas-Electric Coupling: Attacks on the gas side can force expensive changes on the electricity side (and vice versa). For example, manipulating gas load data can force gas-fired generators to shut down, leading to massive power-side load shedding.

Key Results

  • Economic Impact: In the 6-node test case, a successful LR attack increased system costs by 18.2%, primarily by forcing expensive load shedding.
  • Stealthiness vs. Impact: The model successfully identified attacks that stayed within a 50% "forecasting error" margin (making them look like normal fluctuations) while still causing significant disruption.
  • Algorithm Performance: The modified R&D algorithm outperformed standard bilevel solvers (U-R&D), which failed to find the true optimal attack in several scenarios.
  • Scalability: The medium-scale 39-node system was solved in approximately 608 seconds, demonstrating that the approach is viable for near-real-time vulnerability analysis by grid operators.

Limitations and Future Work

  • Complete Information Assumption: The model assumes the attacker has perfect knowledge of the system's topology and parameters. While this represents a "worst-case" scenario, real-world attackers might have incomplete data.
  • Static Modeling: The current work is a "snapshot" in time. Future work could look at multi-period attacks that exploit gas "line pack" (storage within the pipes) or energy storage systems.
  • Defense Strategies: This paper focuses on modeling the attack. The authors suggest that future research should focus on developing automated detection and mitigation strategies based on these findings.

Summary updated

Abstract

As the share of gas-fired power generation continues to increase, it is essential to consider integrated electricity-gas systems (IEGSs). However, the development of IEGSs has led to an increased reliance on cyber infrastructures, such as communication systems, which in turn makes IEGSs more susceptible to cyberattacks. This paper investigates the paradigm and stealthy conditions of load redistribution (LR) attacks on IEGSs and proposes a bilevel mixed-integer model to identify the most severe LR attack from an economic perspective. Under a mild assumption, we prove that the proposed model does not exclude any possible upper-level attack, which differs from existing models that may yield suboptimal solutions. A modified reformulation and decomposition (R&D) algorithm is developed to solve this model in a master-subproblem framework. Particularly, we design a new subproblem to address potential infeasibility issues in the master problem. Accordingly, two types of cuts are added to the master problem for ensuring algorithm feasibility and solution optimality. Testing results validate the effectiveness of the proposed model and solution method.
BibTeX
@article{liu2024modeling,
  author = {Rong-Peng Liu and Xiaozhe Wang and Bo Zeng and Rawad Zgheib},
  title = {{Modeling Load Redistribution Attacks in Integrated Electricity-Gas Systems}},
  journal = {IEEE Trans. Smart Grid},
  year = {2024},
  url = {https://openreview.net/forum?id=jLfE2iTKR1},
}

Loading related content...

Figure