Centralised vs. Decentralised Federated Load Forecasting: Who Holds the Key to Adversarial Attack Robustness?

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

---

Smart Grid Protection with AI and Cryptographic Security

Published: June 5, 2024

Language: Английский

---

Future Trends and Challenges in Cybersecurity and Generative AI

Advances in information security, privacy, and ethics book series, Journal Year: 2024, Volume and Issue: unknown, P. 491 - 522

Published: July 26, 2024

The chapter presents a comprehensive exploration of the changing dynamics at intersection between rapidly growing landscape interconnectivity various devices—the internet things—and innovations piloted by advancements in generative artificial intelligence. In following background-focused analysis, significance enactment new levels security details this fast-growing and virulently expansive is emphasized, with AI ultimately serving as highlight. conversation consequently shifts to threats. This includes detailed depiction cybersecurity threats rooted AI, featuring malicious actors incidents, such increasingly popular phenomenon ransomware-as-a-service mirror illustrations dynamic multifaceted character these class further proceeds more in-depth detail about most contemporary platforms adversarial networks, variational autoencoders, reinforcement learning—all relevant identifying emerging solutions advance strategies cybersecurity. simultaneously conducts an opportunity threat analysis merger regard ethics, regulations, overall touchpoints tactics. concludes call for unity discourse action industry, academia, government stakeholders summary essential cross-disciplinary aspect that must drive narrative confronting overcoming from research. Having presented structure, has allowed coverage major issues opportunities heart cybersecurity-generative combination. Additionally, it provided forum collaborative fortified efforts regarding securing defending uncertainties unpredictable digital store world.

Language: Английский

---

Assessment of Water Hydrochemical Parameters Using Machine Learning Tools

Sustainability, Journal Year: 2025, Volume and Issue: 17(2), P. 497 - 497

Published: Jan. 10, 2025

Access to clean water is a fundamental human need, yet millions of people worldwide still lack access safe drinking water. Traditional quality assessments, though reliable, are typically time-consuming and resource-intensive. This study investigates the application machine learning (ML) techniques for analyzing river in Barnaul area, located on Ob River Altai Krai. The research particularly highlights use Water Quality Index (WQI) as key factor feature engineering. WQI, calculated using Horton model, integrates nine hydrochemical parameters: pH, hardness, solids, chloramines, sulfate, conductivity, organic carbon, trihalomethanes, turbidity. primary objective was demonstrate contribution WQI enhancing predictive performance analysis. A dataset 2465 records analyzed, with missing values parameters (pH, trihalomethanes) addressed imputation via neural network (NN) architectures optimized genetic algorithms (GAs). Models trained without achieved moderate accuracy, but incorporating dramatically improved across all tasks. For trihalomethanes R2 score increased from 0.68 (without WQI) 0.86 (with WQI). Similarly, 0.35 0.74, 0.27 0.69 after including set.

Language: Английский

---

Deep learning and smart energy-based lightweight urban power load forecasting model for sustainable urban growth

Frontiers in Sustainable Cities, Journal Year: 2025, Volume and Issue: 6

Published: Jan. 15, 2025

Introduction Urban power load forecasting is essential for smart grid planning but hindered by data imbalance issues. Traditional single-model approaches fail to address this effectively, while multi-model methods mitigate splitting datasets incur high costs and risk losing shared distribution characteristics. Methods A lightweight urban model (DLUPLF) proposed, enhancing LSTM networks with contrastive loss in short-term sampling, a difference compensation mechanism, feature extraction layer reduce costs. The adjusts predictions learning differences employs dynamic class-center regularization. Its performance was evaluated through parameter tuning comparative analysis. Results DLUPLF demonstrated improved accuracy imbalanced reducing computational It preserved characteristics outperformed traditional efficiency prediction accuracy. Discussion effectively addresses complexity challenges, making it promising solution forecasting. Future work will focus on real-time applications broader systems.

Language: Английский

---

Review of Power Market Optimization Strategies Based on Industrial Load Flexibility

Energies, Journal Year: 2025, Volume and Issue: 18(7), P. 1569 - 1569

Published: March 21, 2025

New power systems, predominantly based on renewable energy, necessitate active load-side management to effectively alleviate the pressures associated with balancing supply-side fluctuations and demand-side energy requirements. Concurrently, as markets continue evolve, both market ancillary services offer valuable guidance for optimal economic dispatch of industrial loads. Although substantial energy-saving potential exists within production processes, their inherent complexity, dynamic nature, mixed continuous–discrete modal characteristics present significant challenges in achieving accurate efficient response. Conversely, ongoing advancement internet techniques lays a robust technical foundation reliable, stable, economically operation new systems large-scale load This paper starts from load, discusses resources advantages disadvantages industry itself, carefully distinguishes participating make decisions. provides comprehensive review intelligent optimization regulation flexibility response systems. Firstly, it synthesizes three prevalent demand strategies (load shedding, shifting, substitution), along regulatory techniques, considering operational various sectors. It then examines trading modeling flexible loads two environments: market. Subsequently, using non-ferrous electrolytic process case study, explores parameters under usage planning. Finally, perspectives market, innovation, stakeholder engagement, highlights unresolved issues insights into future research directions concerning intelligent, digital, market-driven integration flexibility.

Language: Английский

---

A short report on deep learning synergy for decentralized smart grid cybersecurity

Frontiers in Artificial Intelligence, Journal Year: 2025, Volume and Issue: 8

Published: April 9, 2025

Language: Английский

---

AI-Enabled Predictive Analytics in Smart Grids: The Case of Sweden

Complex Systems Informatics and Modeling Quarterly, Journal Year: 2025, Volume and Issue: 42, P. 43 - 62

Published: April 30, 2025

Smart grids (SGs) revolutionize existing power by using a wide range of developing disruptive technologies to generate clean, efficient, and predictable energy. Our study uses an action research method focuses solely on the first two stages process, diagnosis planning, evaluate ways adopt artificial intelligence (AI) applications in SGs for predictive analytics practice. The stage entails conducting systematic literature review AI SGs, highlighting four areas potential analytics: outage prediction, demand response, control coordination, AI-enabled security optimize decision-making, diagnose faults, improve grid stability security. planning step included document analysis devise methods enable practical implementation smart analytics. Finally, we address implementing transparent analytics, followed conclusion future direction. study’s key is that more needed complete taking (implementing solution), evaluation (assessing results), learning (reflecting lessons learned) phases cycle.

Language: Английский

---

Deep Learning Empowered Intermodal Path Optimization in Logistics: Deep Shortest Approach

WSEAS TRANSACTIONS ON BUSINESS AND ECONOMICS, Journal Year: 2025, Volume and Issue: 22, P. 832 - 844

Published: May 2, 2025

This is particularly important in logistics, where path planning critical for adequate transport and distribution processes. That why classical approaches like Dijkstra’s algorithm have been essential, though they are too weak to handle the complications typical of actual logistics networks. To this end, paper proposes a new framework called DeepShortest, which improves optimization process using deep learning methods. DeepShortest uses neural network training flexibility complexity various logistical contexts. Thus, successfully implements within base deliver high result finding shortest most effective paths transporting goods through global chains. In paper, DEEP Define strategy describes how methodologies cast into component approach. addition, real-world case studies substantiate effectiveness advantage compared with previous methods, generally providing stepped-up route performance resource management. an innovative approach solving problems creative solution issues today’s supply chain. With their capacity work areas conditions change often suggest optimal delivery vehicles, presents itself as invaluable that could drastically transform worldwide.

Language: Английский

---

CLEMD, a circuit-level electrical measurements dataset for electrical energy management

Scientific Data, Journal Year: 2024, Volume and Issue: 11(1)

Published: June 6, 2024

Abstract Enhancing energy efficiency in commercial buildings is crucial for reducing consumption. Achieving this goal requires careful monitoring and analysis of the usage patterns exhibited by different devices. Nonetheless, gathering data from individual appliances presents difficulties due to large number appliances, complex installations, costs. This paper Circuits-Level Electrical Measurements Dataset (CLEMD). The measurement was conducted at main switchboard a set distribution boards instead measuring loads. gathered an institutional setting. It consists 42 records vital electrical parameters including voltage, current, frequency, real power, reactive apparent power factor, odd harmonics currents. device deployed were industry-grade had high sampling rate 200 kHz. measurements done over 40-day period, September 16 2023 October 25 2023. CLEMD first Malaysian public dataset on circuit-level electricity consumption offers opportunities research areas such as load disaggregation circuit level, identification, profile forecasting, pattern recognition.

Language: Английский

---