Scholarlink

This paper presents an application of unsupervised approaches to anomaly detection in smart meter log data within Smart Grids, with a specific focus on detecting False Data Injection (FDI) cyber-attacks. The proposed solution leverages the DBSCAN algorithm, selected for its ability to detect clusters of arbitrary shape and isolate noise without prior knowledge of the number of clusters. To validate the model, a realistic and reproducible test dataset was generated by injecting controlled Gaussian noise into a portion of the data. DBSCAN was then trained and evaluated on this dataset, demonstrating robust performance in distinguishing between normal behavior and simulated anomalies. Results achieved include 87% accuracy, 95% precision, and an F1-score of 89%, highlighting the algorithm’s effectiveness in detecting subtle anomalies while minimizing false positives. Parameter tuning, based on k-distance curve analysis, led to the identification of an optimal configuration (eps = 0.25, min samples = 10), which significantly contributed to the quality of clustering and anomaly separation. Overall, the study confirms the viability of unsupervised techniques like DBSCAN in detecting cyber threats and operational issues in Smart Grid environments..

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

False Data Injection Identification in Energy Microgrids Through an Anomaly Detection Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • A. Sgueglia,
  • G. Spinelli,
  • C. A. Visaggio,
  • S. De Vito,
  • G. Di Francia

摘要

This paper presents an application of unsupervised approaches to anomaly detection in smart meter log data within Smart Grids, with a specific focus on detecting False Data Injection (FDI) cyber-attacks. The proposed solution leverages the DBSCAN algorithm, selected for its ability to detect clusters of arbitrary shape and isolate noise without prior knowledge of the number of clusters. To validate the model, a realistic and reproducible test dataset was generated by injecting controlled Gaussian noise into a portion of the data. DBSCAN was then trained and evaluated on this dataset, demonstrating robust performance in distinguishing between normal behavior and simulated anomalies. Results achieved include 87% accuracy, 95% precision, and an F1-score of 89%, highlighting the algorithm’s effectiveness in detecting subtle anomalies while minimizing false positives. Parameter tuning, based on k-distance curve analysis, led to the identification of an optimal configuration (eps = 0.25, min samples = 10), which significantly contributed to the quality of clustering and anomaly separation. Overall, the study confirms the viability of unsupervised techniques like DBSCAN in detecting cyber threats and operational issues in Smart Grid environments..