Scholarlink

<p>Investigating the increasing complexity levels and operational issues of today’s smart grids, this research presents a novel approach based on Transformer-Based Models and Federated Learning for distribution automation and fault tolerance improvement. Transformers are designed for scalable temporal data processing to effectively predict the grid condition, optimize energy consumption, or detect and prevent faults, while FL provides a secure and privacy-preserving collaborative training of Decentralized Regions to develop a highly resilient learning system across agents. These results surpass conventional approaches such as Reinforcement Learning (RL) and Gradient Boosting Machines (GBM), enhance 40% in grid reliability, 30% decrease in average outage time, and a fault detection accuracy rate of 95% or more. These are confirmed through simulative modeling performed on MATLAB, MATPOWER and Python packages, and are characterized by scalability, adaptability and accurate efficiency, suitable for the upcoming generation of smart grid services. This paper proposes a hybrid Transformer-based Federated Learning (T-FL) framework for enhancing reliability and fault detection in smart grid systems. The proposed model captures temporal dependencies in grid data using Transformer architecture while ensuring data privacy through decentralized Federated Learning. Experiments are conducted on the IEEE 30-bus test system using MATPOWER with time-varying load profiles and 25% renewable energy penetration. The dataset consists of 12,000 time-series samples, split into 70% training, 15% validation, and 15% testing. The proposed method achieves 94.8% ± 1.9% fault detection accuracy, reduces outage occurrence by 28.6% ± 2.1%, and improves system reliability by 37.2% ± 2.5%, averaged over 10 independent runs. Comparative evaluation against Reinforcement Learning (RL) and Gradient Boosting Machine (GBM) models demonstrates superior performance under identical training conditions. These results validate the effectiveness of the proposed framework in improving smart grid resilience and operational efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Energy-aware Advanced Machine Learning Models for Reliable Distribution Automation in Future Smart Grids

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • V. Eswaramoorthy,
  • L. Mubaraali,
  • S. SathishKumar,
  • Huda Fatima

摘要

Investigating the increasing complexity levels and operational issues of today’s smart grids, this research presents a novel approach based on Transformer-Based Models and Federated Learning for distribution automation and fault tolerance improvement. Transformers are designed for scalable temporal data processing to effectively predict the grid condition, optimize energy consumption, or detect and prevent faults, while FL provides a secure and privacy-preserving collaborative training of Decentralized Regions to develop a highly resilient learning system across agents. These results surpass conventional approaches such as Reinforcement Learning (RL) and Gradient Boosting Machines (GBM), enhance 40% in grid reliability, 30% decrease in average outage time, and a fault detection accuracy rate of 95% or more. These are confirmed through simulative modeling performed on MATLAB, MATPOWER and Python packages, and are characterized by scalability, adaptability and accurate efficiency, suitable for the upcoming generation of smart grid services. This paper proposes a hybrid Transformer-based Federated Learning (T-FL) framework for enhancing reliability and fault detection in smart grid systems. The proposed model captures temporal dependencies in grid data using Transformer architecture while ensuring data privacy through decentralized Federated Learning. Experiments are conducted on the IEEE 30-bus test system using MATPOWER with time-varying load profiles and 25% renewable energy penetration. The dataset consists of 12,000 time-series samples, split into 70% training, 15% validation, and 15% testing. The proposed method achieves 94.8% ± 1.9% fault detection accuracy, reduces outage occurrence by 28.6% ± 2.1%, and improves system reliability by 37.2% ± 2.5%, averaged over 10 independent runs. Comparative evaluation against Reinforcement Learning (RL) and Gradient Boosting Machine (GBM) models demonstrates superior performance under identical training conditions. These results validate the effectiveness of the proposed framework in improving smart grid resilience and operational efficiency.