TLDR: FO-MADS is a new framework for detecting cyber-attacks and faults in smart microgrids using only one sensor. It employs dual fractional-order derivatives to analyze power signals and a two-stage hierarchical classifier to pinpoint issues. Enhanced by Progressive Memory-Replay Adversarial Training (PMR-AT), it achieves high accuracy (over 92%) even under various cyber-attack scenarios, offering a cost-effective and robust solution for microgrid security.
Smart microgrids, which are localized energy systems, are crucial for modern power distribution. However, their increasing reliance on information and communication technologies makes them vulnerable to cyber-attacks. Traditional methods for diagnosing faults and attacks often require expensive multi-point sensors or rely on assumptions that are not practical in real-world scenarios. This creates a significant challenge for maintaining the safe and reliable operation of these vital energy systems.
A new framework, called the Fractional-Order Memory-Enhanced Attack-Diagnosis Scheme (FO-MADS), has been proposed to address these challenges. This innovative system offers a cost-effective and highly robust solution for detecting cyber-attacks and localizing faults in smart microgrids using data from just a single sensor.
The core innovation of FO-MADS lies in its ability to analyze voltage, active power, and reactive power (VPQ) signals from a single measurement point. It employs a unique dual fractional-order feature extraction method. This involves using two complementary mathematical operators: the Caputo derivative, which is excellent at detecting sudden, high-frequency changes like switch faults, and the Grünwald-Letnikov derivative, which is designed to pick up on subtle, slow-drift anomalies often associated with stealthy cyber-attacks like false data injection. By combining these two approaches, FO-MADS creates a comprehensive and informative feature library from minimal sensor data.
Once the features are extracted, FO-MADS uses a two-stage hierarchical classifier. The first stage identifies which inverter in the microgrid is experiencing an issue, narrowing down the problem from many possibilities to a specific area. If a fault is detected, the second stage then precisely pinpoints the exact faulty switch within that identified inverter. This hierarchical approach not only improves accuracy by managing the complexity of many potential fault classes but also makes the diagnosis more interpretable, telling operators exactly where the problem is.
To ensure the system is resilient against various cyber threats, FO-MADS incorporates a sophisticated training method called Progressive Memory-Replay Adversarial Training (PMR-AT). This method exposes the system to increasingly complex and challenging attack scenarios during training. It uses techniques like Online Hard Example Mining (OHEM) to focus on the most difficult-to-classify samples and replays historical strong attacks to prevent the model from forgetting past threats. This progressive hardening makes FO-MADS exceptionally robust against bias, noise, data replacement, and replay attacks.
Experiments conducted on a four-inverter microgrid testbed, simulating both normal operation and 24 different fault classes under four distinct attack scenarios, demonstrated the effectiveness of FO-MADS. The system achieved diagnostic accuracies of 96.6% under bias attacks, 94.0% under noise attacks, 92.8% under data replacement attacks, and 95.7% under replay attacks. Under attack-free conditions, its accuracy was an impressive 96.7%. These results significantly outperform traditional machine learning models and other diagnostic methods, especially in adversarial environments.
Also Read:
- HERO: A New Framework for Robustness Testing in Industrial AI Systems
- Unpacking Hyperspectral Anomaly Detection: A Comprehensive Review of Methods and Performance
FO-MADS represents a significant step forward in securing smart microgrids. Its ability to achieve high diagnostic fidelity with only a single sensor makes it a cost-effective and easily deployable solution. Its robust design, incorporating advanced feature engineering and adversarial training, markedly enhances the cyber-physical resilience of these critical energy infrastructures. For more details, you can refer to the full research paper: A Single-Point Measurement Framework for Robust Cyber-Attack Diagnosis in Smart Microgrids Using Dual Fractional-Order Feature Analysis.
