Scholarlink

<p>The ongoing proliferation of Internet of Things (IoT) networks in smart environments has heightened the demand for efficient, low-latency, and secure communication across dynamic and resource-constrained infrastructures. To address this challenge, we propose a federated edge intelligence framework that integrates Gated Recurrent Unit (GRU)-based traffic prediction with reinforcement learning (RL)-driven routing, enabling fully decentralized and privacy-preserving adaptive communication. Each IoT node independently trains a GRU model using federated learning to forecast short-term congestion and guide an RL agent in real-time route selection, eliminating the need for centralized control. Differential privacy is incorporated into the aggregation process to safeguard user data. The framework was implemented on edge devices and evaluated through OMNeT++ simulations and both real-world and simulated testbeds. Experimental results demonstrate that our decentralized method outperforms centralized GRU, RL-only, and conventional routing baselines, achieving 91.5% prediction accuracy, over 91% packet delivery ratio, and latencies below 75&#xa0;ms in dense network conditions. These findings indicate that the proposed framework can provide low-latency, privacy-aware communication with reliable data delivery in dense IoT deployments, making it a promising candidate for use in smart city, industrial IoT, and real-time monitoring scenarios.</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

Federated edge intelligence for secure and adaptive routing in IoT: a GRU–RL based framework

  • 期刊论文

  • 发表日期：

    2026年3月7日

查看全文
  • Wajih Abdallah

摘要

The ongoing proliferation of Internet of Things (IoT) networks in smart environments has heightened the demand for efficient, low-latency, and secure communication across dynamic and resource-constrained infrastructures. To address this challenge, we propose a federated edge intelligence framework that integrates Gated Recurrent Unit (GRU)-based traffic prediction with reinforcement learning (RL)-driven routing, enabling fully decentralized and privacy-preserving adaptive communication. Each IoT node independently trains a GRU model using federated learning to forecast short-term congestion and guide an RL agent in real-time route selection, eliminating the need for centralized control. Differential privacy is incorporated into the aggregation process to safeguard user data. The framework was implemented on edge devices and evaluated through OMNeT++ simulations and both real-world and simulated testbeds. Experimental results demonstrate that our decentralized method outperforms centralized GRU, RL-only, and conventional routing baselines, achieving 91.5% prediction accuracy, over 91% packet delivery ratio, and latencies below 75 ms in dense network conditions. These findings indicate that the proposed framework can provide low-latency, privacy-aware communication with reliable data delivery in dense IoT deployments, making it a promising candidate for use in smart city, industrial IoT, and real-time monitoring scenarios.