Scholarlink

<p>The increasing integration of smart grids has introduced vulnerabilities in Advanced Metering Infrastructure (AMI), particularly through False Data Injection (FDI) attacks, which can compromise grid operations. This study develops a machine learning framework for detecting FDI attacks using both simulated (SD) and real-time (RD) datasets. Three types of FDI attacks—constant offset, percentage change, and random noise—were modeled using key parameters such as voltage, current, and power. Six machine learning algorithms—Random Forest, Gradient Boosting, XGBoost, AdaBoost, K-Nearest Neighbors (KNN), and LightGBM—were evaluated based on accuracy, precision, Recall, F1-score, and ROC curve analysis, with an additional robustness analysis conducted to assess performance consistency under varying conditions. On simulated data (SD), Random Forest achieved the highest accuracy of 0.77 on RD, it outperformed others with an accuracy of 0.94. Gradient Boosting and LightGBM performed strongly in both datasets, achieving accuracies of 0.73 (SD) and 0.96–0.97 (RD), respectively. Robustness analysis indicated that Random Forest and XGBoost maintained the highest performance consistency under hyperparameter variations, while AdaBoost and KNN exhibited lower accuracy, particularly in detecting random noise attacks. These results demonstrate the effectiveness of ensemble methods like Random Forest in improving the detection of sophisticated cyberattacks and enhancing AMI systems’ security and reliability.</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

AI-based False Data Injection attack detection in smart grid Advanced Metering Infrastructure

  • 期刊论文

  • 发表日期：

    2025年7月27日

查看全文
  • Ramachandran Murugesan,
  • Prabhakar Gunasekaran,
  • Shanthi Jeyabal,
  • Vignesh Kumar Vethanayagam

摘要

The increasing integration of smart grids has introduced vulnerabilities in Advanced Metering Infrastructure (AMI), particularly through False Data Injection (FDI) attacks, which can compromise grid operations. This study develops a machine learning framework for detecting FDI attacks using both simulated (SD) and real-time (RD) datasets. Three types of FDI attacks—constant offset, percentage change, and random noise—were modeled using key parameters such as voltage, current, and power. Six machine learning algorithms—Random Forest, Gradient Boosting, XGBoost, AdaBoost, K-Nearest Neighbors (KNN), and LightGBM—were evaluated based on accuracy, precision, Recall, F1-score, and ROC curve analysis, with an additional robustness analysis conducted to assess performance consistency under varying conditions. On simulated data (SD), Random Forest achieved the highest accuracy of 0.77 on RD, it outperformed others with an accuracy of 0.94. Gradient Boosting and LightGBM performed strongly in both datasets, achieving accuracies of 0.73 (SD) and 0.96–0.97 (RD), respectively. Robustness analysis indicated that Random Forest and XGBoost maintained the highest performance consistency under hyperparameter variations, while AdaBoost and KNN exhibited lower accuracy, particularly in detecting random noise attacks. These results demonstrate the effectiveness of ensemble methods like Random Forest in improving the detection of sophisticated cyberattacks and enhancing AMI systems’ security and reliability.