[Paper Review] Vulnerability Assessment of N-1 Reliable Power Systems to False Data Injection Attacks

Zhigang Chu, Jiazi Zhang+2 more|arXiv (Cornell University)|Mar 19, 2019

Smart Grid Security and Resilience17 references4 citations

TL;DR

This paper proposes an attacker-defender bi-level linear program (ADBLP) to assess the vulnerability of $N-1$ reliable power systems to false data injection (FDI) attacks that maximize post-contingency line flow. By modeling the system operator's response via security-constrained economic dispatch (SCED) and using a modified Benders' decomposition algorithm, the study demonstrates that intelligent FDI attacks can induce physical overflows on target transmission lines while evading detection, with simulations on a 2000-bus Texas system showing up to 112.2% overflow on a target line under contingency.

ABSTRACT

This paper studies the vulnerability of large-scale power systems to false data injection (FDI) attacks through their physical consequences. Prior work has shown that an attacker-defender bi-level linear program (ADBLP) can be used to determine the worst-case consequences of FDI attacks aiming to maximize the physical power flow on a target line. Understanding the consequences of these attacks requires consideration of power system operations commonly used in practice, specifically real-time contingency analysis (RTCA) and security constrained economic dispatch (SCED). An ADBLP is formulated with detailed assumptions on attacker's knowledge, and a modified Benders' decomposition algorithm is introduced to solve such an ADBLP. The vulnerability analysis results presented for the synthetic Texas system with 2000 buses show that intelligent FDI attacks can cause post-contingency overflows.

Motivation & Objective

• To assess the physical vulnerability of large-scale $N-1$ reliable power systems to false data injection (FDI) attacks that exploit real-time contingency analysis (RTCA) and security-constrained economic dispatch (SCED).
• To model the attacker’s knowledge and constraints required to launch unobservable FDI attacks that bypass bad data detectors and manipulate system state estimation.
• To develop an efficient solution method for the non-convex attacker-defender bi-level linear program (ADBLP) that models the attacker’s goal and the system operator’s SCED response.
• To quantify the worst-case physical consequences of such attacks, including post-contingency line overflows and unintended violations on non-targeted branches.

Proposed method

• Formulates an attacker-defender bi-level linear program (ADBLP) where the upper level models the attacker’s objective to maximize power flow on a target line, and the lower level models the system operator’s SCED response under RTCA-generated security constraints.
• Introduces detailed assumptions on attacker knowledge, including access to system topology, line limits, contingency scenarios, and the ability to observe or predict RTCA and SCED outcomes.
• Reformulates the ADBLP using Karush-Kuhn-Tucker (KKT) conditions to transform it into a mathematical program with equilibrium constraints (MPEC), then further reformulates it as a mixed-integer linear program (MILP) via big-M reformulation of complementary slackness.
• Proposes a modified Benders’ decomposition algorithm to efficiently solve the large-scale ADBLP, decomposing the problem into master and subproblems to handle the bi-level structure.
• Employs $l_1$-norm and $l_0$-norm constraints on the attack vector to model detectability and sparsity, linking attack magnitude to load redistribution detectability.
• Validates the attack model and solution method on a synthetic 2000-bus Texas power system, simulating RTCA and SCED responses under injected false measurements.

Experimental results

Research questions

• RQ1 What level of knowledge does an attacker need to design unobservable FDI attacks that successfully manipulate SCED and induce physical line overflows in $N-1$ reliable systems?
• RQ2 How can the worst-case FDI attack be modeled as a bi-level optimization problem where the attacker maximizes target line flow and the system operator responds via SCED?
• RQ3 What is the physical impact of such attacks, including post-contingency overflows and unintended violations on non-targeted branches?
• RQ4 How can the non-convex and computationally challenging ADBLP be efficiently solved for large-scale systems like the 2000-bus Texas model?
• RQ5 To what extent can detectability of FDI attacks be quantified using load shift and attack vector sparsity metrics?

Key findings

• The proposed ADBLP-based attack design successfully induced post-contingency overflows on 8 out of 25 target branches in the Texas system, with the highest physical overflow reaching 112.2% on line ln-2025-2055 under contingency ln-2054-5236.
• Even though the attacker targeted only one line, the attacks caused unintended violations on four additional branches, indicating cascading physical consequences beyond the target.
• The attack on line ln-2025-2055 achieved a 112.2% post-contingency power flow, demonstrating that $N-1$ security constraints can be circumvented by intelligent FDI attacks.
• The $l_0$-norm of the attack vector ranged from 90 to 461 across successful attacks, indicating that only a small number of measurements need to be manipulated to cause significant physical damage.
• The system operator observed no violations in the cyber state (i.e., the system appeared secure), while the physical system experienced severe overflows, confirming the attack's unobservability.
• The study identifies specific lines, such as ln-6188-7305 and ln-7233-7251, as highly vulnerable, with physical power flows exceeding 105% under attack, suggesting they require enhanced protection.