Scholarlink

When it comes to cyber-physical systems (CPS) such as energy-aware smart homes (EASH), accurate differentiation between equipment failures and cyber threats is a critical aspect of operational safety. The proposed method collects sensor data when the system is running normally and then performs data cleansing to set up patterns. Feature engineering acquires essential attributes and time fragmentation. Compiling these features along with real-time communication data, the system provides a training for long short-term memory (LSTM) network of pretrained for the analysis of sequential information. The inclusion of historical trends to the stronger and more flexible structure of the improved framework enhances this objective by providing flexibility to identify the type of anomaly within the EASH system, with the potential to achieve peak accuracy of 88.5% compared to existing methods.

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

Network Anomaly Classification Using Long Short-Term Memory Networks in Cyber-Physical Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • V. Saravanan,
  • Abhishek Awasthi,
  • T. Sathis Kumar,
  • Tanneeru Sudha Rani,
  • E. G. Satish

摘要

When it comes to cyber-physical systems (CPS) such as energy-aware smart homes (EASH), accurate differentiation between equipment failures and cyber threats is a critical aspect of operational safety. The proposed method collects sensor data when the system is running normally and then performs data cleansing to set up patterns. Feature engineering acquires essential attributes and time fragmentation. Compiling these features along with real-time communication data, the system provides a training for long short-term memory (LSTM) network of pretrained for the analysis of sequential information. The inclusion of historical trends to the stronger and more flexible structure of the improved framework enhances this objective by providing flexibility to identify the type of anomaly within the EASH system, with the potential to achieve peak accuracy of 88.5% compared to existing methods.