Scholarlink

The rapid expansion of Industrial Internet of Things (IIoT) networks has increased the risk of sophisticated cyberattacks, necessitating accurate and adaptive anomaly detection methods. This paper proposes a hybrid Fuzzy-Long Short-Term Memory (Fuzzy-LSTM) framework that combines the temporal modeling capabilities of LSTM networks with an adaptive fuzzy inference system for dynamic threshold adjustment. The objective of the approach is to eliminate the problems associated with commonly used static thresholds in traditional anomaly detection systems so that false alarms can be reduced while retaining high sensitivity of detection. The model was compared with Autoencoder, CNN, and traditional LSTM-based techniques using two benchmark datasets, IoT23, and WUSTL-IIoT-2021. According to the experimental results, Fuzzy-LSTM outperformed all baselines in terms of accuracy, precision, and recall, reaching the greatest performance on IoT23 with an accuracy of up to 96.23% and an F1-score of 98%. This demonstrated the framework’s resilience and adaptability in identifying a wide range of attack behaviors in IIoT contexts. With applications for operational safety, intrusion detection, and predictive maintenance, the suggested approach presents a viable route for the deployment of intelligent, real-time anomaly detection in the industrial sector.

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

Anomaly Detection in IoT Sensor Data Using Deep Learning for Industrial IoT

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hussein M. Farhood,
  • Zaid J. Al-Araji,
  • Sara Raad Qasim,
  • Ameer M. Al-Obaidi,
  • Fahad Ahmed Shaban,
  • Zaid Ali Abdulkadhim,
  • Khalid M. Farhood,
  • Husam Nawfel Fadhil

摘要

The rapid expansion of Industrial Internet of Things (IIoT) networks has increased the risk of sophisticated cyberattacks, necessitating accurate and adaptive anomaly detection methods. This paper proposes a hybrid Fuzzy-Long Short-Term Memory (Fuzzy-LSTM) framework that combines the temporal modeling capabilities of LSTM networks with an adaptive fuzzy inference system for dynamic threshold adjustment. The objective of the approach is to eliminate the problems associated with commonly used static thresholds in traditional anomaly detection systems so that false alarms can be reduced while retaining high sensitivity of detection. The model was compared with Autoencoder, CNN, and traditional LSTM-based techniques using two benchmark datasets, IoT23, and WUSTL-IIoT-2021. According to the experimental results, Fuzzy-LSTM outperformed all baselines in terms of accuracy, precision, and recall, reaching the greatest performance on IoT23 with an accuracy of up to 96.23% and an F1-score of 98%. This demonstrated the framework’s resilience and adaptability in identifying a wide range of attack behaviors in IIoT contexts. With applications for operational safety, intrusion detection, and predictive maintenance, the suggested approach presents a viable route for the deployment of intelligent, real-time anomaly detection in the industrial sector.