Scholarlink

<p>This paper presents an innovative framework designed to enhance the security of agricultural Internet of Things (IoT) systems. The Secure Federated Deep Reinforcement Learning Intrusion Detection System (SFEDRL-IDS) is a highly efficient solution that leverages Federated Double Deep Q-learning (FDDQN) to collectively train a network intrusion detection system (NIDS) in a distributed manner. The confidentiality of local model weights is safeguarded through the application of Fully Homomorphic Encryption (FHE) and Cryptographic Keys over Polynomial Rings (CKKS). To effectively mitigate IoT attacks, the FELIDS system employs three distinct Deep Learning (DL) classifiers: Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Long Short-Term Memory networks (LSTM), which collaboratively counteract threats targeting agricultural IoT environments. The results not only demonstrate the superiority of the SFEDRL-IDS system compared to traditional centralized machine learning techniques but also highlight its effectiveness in preserving the privacy of agricultural IoT systems. The SFEDRL-IDS system achieves remarkable accuracy rates of 98.67% for multiclass classifiers and 97.73% for binary classifiers. Furthermore, its minimal CPU and memory resource utilization attests to its efficiency. Additionally, we deploy the proposed framework within an intelligent irrigation system, with results further validating its effectiveness.</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

Sfedrl-ids: secure federated deep reinforcement learning-based intrusion detection system for agricultural internet of things

  • 期刊论文

  • 发表日期：

    2025年6月16日

查看全文
  • Rabaie Benameur,
  • Amine Dahane

摘要

This paper presents an innovative framework designed to enhance the security of agricultural Internet of Things (IoT) systems. The Secure Federated Deep Reinforcement Learning Intrusion Detection System (SFEDRL-IDS) is a highly efficient solution that leverages Federated Double Deep Q-learning (FDDQN) to collectively train a network intrusion detection system (NIDS) in a distributed manner. The confidentiality of local model weights is safeguarded through the application of Fully Homomorphic Encryption (FHE) and Cryptographic Keys over Polynomial Rings (CKKS). To effectively mitigate IoT attacks, the FELIDS system employs three distinct Deep Learning (DL) classifiers: Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Long Short-Term Memory networks (LSTM), which collaboratively counteract threats targeting agricultural IoT environments. The results not only demonstrate the superiority of the SFEDRL-IDS system compared to traditional centralized machine learning techniques but also highlight its effectiveness in preserving the privacy of agricultural IoT systems. The SFEDRL-IDS system achieves remarkable accuracy rates of 98.67% for multiclass classifiers and 97.73% for binary classifiers. Furthermore, its minimal CPU and memory resource utilization attests to its efficiency. Additionally, we deploy the proposed framework within an intelligent irrigation system, with results further validating its effectiveness.