Novel Evasion Attacks Against Adversarial Training Defense for Smart Grid Federated Learning DOI Creative Commons
Atef H. Bondok, Mohamed Mahmoud, Mahmoud M. Badr

et al.

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 112953 - 112972

Published: Jan. 1, 2023

In the advanced metering infrastructure (AMI) of smart grid, meters (SMs) are deployed to collect fine-grained electricity consumption data, enabling billing, load monitoring, and efficient energy management. However, some consumers engage in fraudulent behavior by hacking their meters, leading either traditional theft or more sophisticated evasion attacks (EAs). EAs aim illegally reduce bills while deceiving detection mechanisms. The current methods for identifying such raise privacy concerns due need access consumers' detailed data train To address concerns, federated learning (FL) is proposed as a collaborative training approach across multiple consumers. Adversarial (AT) has shown promise countering threats on machine models. This paper, first, investigates susceptibility classifiers trained FL both independent identically distributed (IID) Non-IID data. Then, it effectiveness AT securing global detector against EAs, assuming no misbehavior from participant process. After that, we introduce three novel attacks, namely Distillation , xmlns:xlink="http://www.w3.org/1999/xlink">No-Adversarial-Sample-Training xmlns:xlink="http://www.w3.org/1999/xlink">False-Labeling which can be launched during process make model susceptible at inference time. Finally, extensive experiments conducted validate severity these attacks. Our findings reveal that counter effectively when participants honest, but fails they act maliciously launch our works lays foundation future endeavors exploring additional countermeasures, conjunction with AT, bolster security resilience models adversarial context detection.

Language: Английский

An Adaptive Deep Learning Neural Network Model to Enhance Machine-Learning-Based Classifiers for Intrusion Detection in Smart Grids DOI Creative Commons
Xue Jun Li, Maode Ma,

Yihan Sun

et al.

Algorithms, Journal Year: 2023, Volume and Issue: 16(6), P. 288 - 288

Published: June 2, 2023

Modern smart grids are built based on top of advanced computing and networking technologies, where condition monitoring relies secure cyberphysical connectivity. Over the network infrastructure, transported data containing confidential information, must be protected as vulnerable subject to various cyberattacks. Various machine learning classifiers were proposed for intrusion detection in grids. However, each them has respective advantage disadvantages. Aiming improve performance existing classifiers, this paper proposes an adaptive deep algorithm with a pre-processing module, neural pre-training module classifier which work together classify types using their high-dimensional features. The Adaptive Deep Learning (ADL) obtains number layers neurons per layer by determining characteristic dimension traffic. With transfer learning, ADL can extract original dimensions obtain new abstract By combining models traditional learning-based classification models, traffic is significantly improved. Network Security Laboratory-Knowledge Discovery Databases (NSL-KDD) dataset, experimental results show that improves effectiveness methods reduces training time, indicating promising candidate enhance security

Language: Английский

Citations

13
Deep Anomaly Detection Framework Utilizing Federated Learning for Electricity Theft Zero-Day Cyberattacks DOI Creative Commons
Ali Alshehri, Mahmoud M. Badr, Mohamed Baza

et al.

Sensors, Journal Year: 2024, Volume and Issue: 24(10), P. 3236 - 3236

Published: May 20, 2024

Smart power grids suffer from electricity theft cyber-attacks, where malicious consumers compromise their smart meters (SMs) to downscale the reported consumption readings. This problem costs electric utility companies worldwide considerable financial burdens and threatens grid stability. Therefore, several machine learning (ML)-based solutions have been proposed detect theft; however, they limitations. First, most existing works employ supervised that requires availability of labeled datasets benign usage samples. Unfortunately, this approach is not practical due scarcity real Moreover, training a detector on specific cyberattack scenarios results in robust against those attacks, but it might fail new attack scenarios. Second, although few investigated anomaly detectors for theft, none addressed consumers’ privacy. To address these limitations, paper, we propose comprehensive federated (FL)-based deep detection framework tailored practical, reliable, privacy-preserving energy detection. In our framework, train local autoencoder-based private data only share trained detectors’ parameters with an EUC aggregation server iteratively build global detector. Our extensive experimental demonstrate superior performance compared also capability FL-based accurately zero-day attacks while preserving

Language: Английский

Citations

5
A contemporary survey of recent advances in federated learning: Taxonomies, applications, and challenges DOI
Mohammed H. Alsharif, Raju Kannadasan, Wei Wei

et al.

Internet of Things, Journal Year: 2024, Volume and Issue: 27, P. 101251 - 101251

Published: June 15, 2024

Language: Английский

Citations

5
Empowering Smart Grid Security: Towards Federated Learning in 6G-Enabled Smart Grids using Cloud DOI Creative Commons

J. Jithish,

Nagarajan Mahalingam, Kiat Seng Yeo

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 11, 2024

Abstract As the realm of smart grids continues to evolve, embracing new technologies, researchers are exploring potential ofupcoming 6G technology address challenges in management grids. With adoption technology,wireless energy meters, which play a key role grid advancement, promise higher data rates, ultra-low latency, improvedconnectivity, and enhanced security. However, integration advanced technologies into grids, raises concernregarding cyberattacks such as distributed-denial-of-service (DDoS) attacks, pose grave threat functionality andstability grid. To these security challenges, traditionally implement intrusion detection systems (IDS)that analyse traffic logs from but traditional IDS may face difficulties detecting novel attacks subtle,multi-domain DDoS attacks. Towards securing anomaly emerges crucial technique, integratedwith deep learning (DL), this technique can potentially identify deviations normal, non-malicious network traffic, detectcyberattacks, thereby enhancing it is seen that using user for training DL models at serverviolates privacy regulations, necessitates balance between strict adherence todata norms. Federated Learning (FL) has emerged suitable solution scenario, offering privacy-focusedsolution allowing meters train with locally generated datasets make predictions edge. In thiswork, we propose hierarchical FL approach era, focusing on privacy-preserving detectionagainst Our work integrates cloud-based service framework within an setup leveraging thescalability cloud platforms edge computing efficient, secure, cost-effective line 6Gtechnology requirements. Evaluation our local simulation environment, workstation serverand Raspberry Pi devices client nodes infrastructure provided by Amazon Web Services (AWS). goal toinvestigate feasibility solutions support federated learning-based grids.Theperformance metrics simulations custom neural showed variations betweentwo sets not significant, proposed deployment real-world scenarios,especially upcoming 6G-enabled where consistent performance essential.

Language: Английский

Citations

4
Centralised vs. Decentralised Federated Load Forecasting: Who Holds the Key to Adversarial Attack Robustness? DOI Creative Commons
Habib Ullah Manzoor, Sajjad Hussain, David Flynn

et al.

Published: June 7, 2024

The integration of AI and ML in energy forecasting is pivotal for modern management. Federated Learning (FL) stands out by enhancing data privacy collaboration among distributed resources, enabling model training while reducing reliance on centralized servers transfers. Despite its merits, FL faces substantial security challenges, particularly from adversarial attacks that can compromise the integrity reliability models. This paper aims to address these concerns examining efficiency Centralized (CFL) Decentralized (DFL) load forecasting. Through comparative analysis utilizing publicly available household datasets short-term forecasting, our study reveals DFL demonstrates superior resilience against compared CFL. Notably, findings indicate impact poisoning confined targeted client DFL, CFL exhibits broader susceptibility across all clients. When attacked, CFL's averaged Mean Absolute Error (MAE) increased 0.076 0.22 kWh, whereas maintained a lower MAE 0.116 kWh. Additionally, we present Random Layer Aggregation (DRLA) augment DFL's robustness, offering further insights into methodologies within contexts.

Language: Английский

Citations

4
Enhancing cyber security in energy hubs with electrical and thermal storage: Leveraging machine learning for demand-supply structure perspective in threat detection DOI

Daryoush Tavangar Rizi,

Mohammad Hassan Nazari, Seyed Hossein Hosseinian

et al.

Journal of Energy Storage, Journal Year: 2025, Volume and Issue: 111, P. 115342 - 115342

Published: Jan. 15, 2025

Language: Английский

Citations

0
Smart Grid Anomaly Detection Using MFDA and Dilated GRU-based Neural Networks DOI Creative Commons

M. Ravinder,

Vikram Kulkarni

Smart Grids and Sustainable Energy, Journal Year: 2025, Volume and Issue: 10(1)

Published: Jan. 16, 2025

Language: Английский

Citations

0
Adversarial Measurements for Convolutional Neural Network-based Energy Theft Detection Model in Smart Grid DOI Creative Commons

Santosh Nirmal,

Pramod Patil, Sagar Shinde

et al.

e-Prime - Advances in Electrical Engineering Electronics and Energy, Journal Year: 2025, Volume and Issue: unknown, P. 100909 - 100909

Published: Jan. 1, 2025

Language: Английский

Citations

0
Federated Learning and Neural Circuit Policies: A Novel Framework for Anomaly Detection in Energy-Intensive Machinery DOI Creative Commons
Giulia Palma, Giovanni Geraci, Antonio Rizzo

et al.

Energies, Journal Year: 2025, Volume and Issue: 18(4), P. 936 - 936

Published: Feb. 15, 2025

In the realm of predictive maintenance for energy-intensive machinery, effective anomaly detection is crucial minimizing downtime and optimizing operational efficiency. This paper introduces a novel approach that integrates federated learning (FL) with Neural Circuit Policies (NCPs) to enhance in compressors utilized leather tanning operations. Unlike traditional Long Short-Term Memory (LSTM) networks, which rely heavily on historical data patterns often struggle generalization, NCPs incorporate physical constraints system dynamics, resulting superior performance. Our comparative analysis reveals significantly outperform LSTMs accuracy interpretability within framework. innovative combination not only addresses pressing privacy concerns but also facilitates collaborative across decentralized sources. By showcasing effectiveness FL NCPs, this research paves way advanced strategies prioritize both performance integrity industries.

Language: Английский

Citations

0
Investigating the Efficiency of a Federated Learning-Based Intrusion Detection System for Smart Grid DOI Creative Commons

Najet Hamdi

SN Computer Science, Journal Year: 2025, Volume and Issue: 6(3)

Published: March 1, 2025

Language: Английский

Citations

0