FaultGuard

A Robust Framework for Resilient Fault Prediction in Smart Electrical Grids
Paper Available »

Emad Efatinasab · Francesco Marchiori · Alessandro Brighente · Mirco Rampazzo · Mauro Conti

Table of Contents

1. Abstract
2. Citation
3. Usage
4. Models
5. Baseline
6. Attacks

🧩 Abstract

Ensuring the reliability of energy distribution networks is paramount for uninterrupted provision, a challenge further compounded by the introduction of new sustainable energy sources to the grid. Promptly addressing faults becomes critical for optimal energy consumption, quick restoration, and overall customer satisfaction in this evolving landscape. While Machine Learning approaches have been applied in smart grids for fault detection, the robustness and security of these systems need thorough exploration. The vulnerability of these systems to adversarial attacks underscores the need for a rigorous examination of their resilience and the implementation of robust security measures. To address these concerns, we introduce FaultGuard, a resilient framework for fault type and zone classification tasks. To ensure the security of our system, we employ an Anomaly Detection System (ADS) leveraging a novel Generative Adversarial Network training layer to identify attacks. Furthermore, we propose a low-complexity fault prediction model and an online adversarial training technique to enhance robustness. We comprehensively evaluate the framework's performance against various adversarial attacks using the publicly available IEEE13-AdvAttack dataset, a simulated dataset based on the IEEE-13 test node feeder. Our model outclasses the state-of-the-art with an accuracy of up to 0.958, and our ADS shows attack detection capabilities with an accuracy of up to 1.000. Moreover, we demonstrate how our novel training layers drastically increase performances across the whole framework, with a mean increase of 154.18% in ADS accuracy and 118.14% in model accuracy. With our research, we contribute to advancing reliable and secure smart grid applications, ensuring uninterrupted and secure energy delivery to consumers, especially in the face of challenges posed by integrating new sustainable energy sources.

(back to top)

🗣️ Citation

Please, cite this work when referring to FaultGuard.

@InProceedings{10.1007/978-3-031-64171-8_26,
author="Efatinasab, Emad
and Marchiori, Francesco
and Brighente, Alessandro
and Rampazzo, Mirco
and Conti, Mauro",
editor="Maggi, Federico
and Egele, Manuel
and Payer, Mathias
and Carminati, Michele",
title="FaultGuard: A Generative Approach to Resilient Fault Prediction in Smart Electrical Grids",
booktitle="Detection of Intrusions and Malware, and Vulnerability Assessment",
year="2024",
publisher="Springer Nature Switzerland",
address="Cham",
pages="503--524",
abstract="Predicting and classifying faults in electricity networks is crucial for uninterrupted provision and keeping maintenance costs at a minimum. Thanks to the advancements in the field provided by the smart grid, several data-driven approaches have been proposed in the literature to tackle fault prediction tasks. Implementing these systems brought several improvements, such as optimal energy consumption and quick restoration. Thus, they have become an essential component of the smart grid. However, the robustness and security of these systems against adversarial attacks have not yet been extensively investigated. These attacks can impair the whole grid and cause additional damage to the infrastructure, deceiving fault detection systems and disrupting restoration.",
isbn="978-3-031-64171-8"
}

(back to top)

⚙️ Usage

To execute the attacks or to deploy the FaultGuard framework, start by cloning the repository:

git clone https://github.com/emadef1/FaultGuard.git
cd FaultGuard

Then, install the required Python packages by running:

pip install -r requirements.txt

(back to top)

🤖 Models

The architecture and training details of our GRU model are extensively expounded in Section 6.2 of the paper. Also, we have chosen to integrate classical machine learning algorithms such as XGBoost, Random Forest, and Decision Tree into our analysis for two primary reasons. Firstly, they serve as a baseline for comparison against our proposed models, enabling us to gauge the performance and efficacy of our approaches. Secondly, their inclusion underscores the significance of considering causality between data points in this task, highlighting the importance of leveraging advanced techniques to capture temporal dependencies within the data.

(back to top)

📊 Baseline

The obtained results are the following.

Model	Accuracy Fault Type	Accuracy Fault Zone
Ardito et al.	0.460	0.710
Decision Tree	0.522	0.818
Random Forest	0.543	0.831
XGBoost	0.560	0.841
GRU	0.604	0.958

(back to top)

⚔️ Attacks

Attacks are divided in different scenarios depending on the level of knowledge that the attacker has of the system.

Scenario	Model	Data
White Box	✅	✅
Gray Box	❌	✅

(back to top)