Abstract
This paper addresses a set-theoretic method for the detection of data corruption cyber-attacks on the load frequency control loop of a networked power system. The system consists of several interconnected control areas forming a power grid. Based on the overall discrete-time network dynamics, a convex and compact polyhedral robust invariant set is extracted and is used as a set-induced anomaly detector. If the state vector exits the invariant set, then an alarm will be activated, and the potential threat is considered disclosed. The attack scenario used to assess the efficiency of the proposed anomaly detector concerns corrupted frequency sensor measurements transmitted to the automatic generation control unit of a compromised control area. Simulation studies highlight the ability of a set-theoretic approach to disclose persistent and intermittent attack patterns even when they occur at the same time with changes in the power load demand.
1. Introduction
Modern power grids are quintessential Cyber-Physical Systems (CPS), integrating digital communication and control networks with physical power generation hardware. A critical component of this infrastructure is the Load Frequency Control (LFC) loop, which maintains grid stability by balancing generation and demand to keep frequency at nominal levels (e.g., 50/60 Hz). However, this digital integration introduces vulnerabilities to False Data Injection (FDI) attacks. Adversaries can corrupt sensor measurements to manipulate the Automatic Generation Control (AGC) unit, potentially leading to grid instability.
Traditional anomaly detection methods often rely on residuals—the difference between observed sensor data and model-based predictions. While effective against simple attacks, sophisticated adversaries can employ intermittent or oscillating corruption patterns to keep residuals below alarm thresholds. This paper proposes a set-theoretic approach to create a robust "safety bubble" that identifies such stealthy maneuvers.
2. Methodology
The proposed detection framework utilizes a Robust Invariant Set (RIS) to monitor system integrity.
2.1 System Modeling and State Augmentation
The researchers developed a discrete-time state-space model representing a networked power system with multiple interconnected control areas. To enhance detection capabilities, the model incorporates physical states (frequency deviation, mechanical power) and augmented states representing the accumulated time errors from the system's integral controllers.
2.2 Robust Invariant Set Extraction
The core of the detector is a convex, compact polyhedral set derived from the system's dynamics and known bounds on routine disturbances (e.g., standard load fluctuations). Mathematically, for a healthy system operating within expected disturbance bounds, the state vector must remain within this RIS indefinitely.
2.3 Anomaly Detection Logic
The detector operates as a binary classifier:
- Normal Operation: If the state vector resides within the RIS, the alarm signal remains at 0.
- Anomaly Detected: If the state vector exits the polyhedral bounds, the alarm signal switches to 1, indicating a potential cyber-attack.
3. Experimental Setup
The methodology was validated using a benchmark two-area power plant model, a standard in power systems research. The simulation environment accounts for power base, inertia constants, and other critical parameters to ensure a realistic representation of grid dynamics. To test the detector's robustness, the authors simulated "hysteresis-based switching" attacks, where corruption is toggled to evade detection.
4. Results and Discussion
4.1 Detection Performance
Simulation results demonstrated high sensitivity. A 4.5 Hz bias attack was detected almost instantaneously. Even lower-magnitude attacks (e.g., 2 Hz), which might bypass conventional threshold filters, were successfully disclosed as the system state drifted outside the RIS.
4.2 Exploiting Integral Control
A key insight is that the system's own stabilization mechanism—integral control—aids in detection. While an attacker might mask immediate frequency spikes, they cannot easily hide the linear divergence of augmented error states. This "drift" eventually forces the state vector out of the invariant set.
4.3 Resilience to Load Noise
The set-theoretic approach effectively distinguishes between legitimate power load changes and malicious data corruption. By incorporating disturbance bounds into the RIS definition, the system avoids the "false positive" traps common in simpler monitoring tools.
5. Conclusion
The set-theoretic anomaly detector provides a mathematically rigorous method for securing power grid LFC loops. Its primary advantage lies in its ability to detect stealthy, low-magnitude, or intermittent attacks by monitoring the system's position in multi-dimensional state space rather than just signal magnitude.
Future research will focus on optimizing the computational complexity of RIS extraction for large-scale national grids and improving resilience to coordinated attacks across multiple control areas. Additionally, addressing model dependency remains a priority, as the invariant set must be updated to reflect physical changes in grid topology.
Summary updated