Scholarlink

<p>With the rapid development of the Industrial Internet of Things (IIoT), it plays a crucial role in fields such as smart manufacturing, intelligent sensing, and automated control. However, due to the high real-time requirements, strong security sensitivity, and device heterogeneity in IIoT environments, frequent cross-domain data interactions have triggered a series of challenges in data sharing security and privacy protection. These include data silos, parameter privacy leakage, insufficient incentives, and poor adaptability of consensus mechanisms. To address these issues, this paper proposes a Blockchain-based Trusted Federated Learning scheme for Industrial Internet of Things (BTFL). First, the scheme integrates direct trust and indirect trust to compute comprehensive credibility for nodes, with the results stored on the blockchain. Based on this, a trust-driven improved PBFT consensus mechanism (IM-PBFT) is designed. This mechanism dynamically selects consensus and validation nodes, enhancing system robustness and consensus efficiency. Second, a dynamic token incentive mechanism is introduced to reward high-trust nodes and penalize malicious ones, fostering honest participation and sustained collaboration. Finally, to prevent model parameter leakage and inference attacks, an adaptive differential privacy mechanism based on cross-entropy feedback is proposed. This mechanism dynamically adjusts privacy budgets and noise magnitude in real-time according to the model’s convergence state, achieving a balance between privacy protection and model accuracy. Experimental results demonstrate that the proposed scheme achieves high global model accuracy on both the MNIST and CIFAR-10 datasets, effectively resisting up to 40<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\%\)</EquationSource> </InlineEquation> model poisoning attacks and confirming its security and reliability.</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

Btfl: Blockchain-based trustworthy federated learning scheme for industrial internet of things

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Hui Wang,
  • Dandan Liu,
  • Zihao Shen,
  • Peiqian Liu,
  • Kun Liu

摘要

With the rapid development of the Industrial Internet of Things (IIoT), it plays a crucial role in fields such as smart manufacturing, intelligent sensing, and automated control. However, due to the high real-time requirements, strong security sensitivity, and device heterogeneity in IIoT environments, frequent cross-domain data interactions have triggered a series of challenges in data sharing security and privacy protection. These include data silos, parameter privacy leakage, insufficient incentives, and poor adaptability of consensus mechanisms. To address these issues, this paper proposes a Blockchain-based Trusted Federated Learning scheme for Industrial Internet of Things (BTFL). First, the scheme integrates direct trust and indirect trust to compute comprehensive credibility for nodes, with the results stored on the blockchain. Based on this, a trust-driven improved PBFT consensus mechanism (IM-PBFT) is designed. This mechanism dynamically selects consensus and validation nodes, enhancing system robustness and consensus efficiency. Second, a dynamic token incentive mechanism is introduced to reward high-trust nodes and penalize malicious ones, fostering honest participation and sustained collaboration. Finally, to prevent model parameter leakage and inference attacks, an adaptive differential privacy mechanism based on cross-entropy feedback is proposed. This mechanism dynamically adjusts privacy budgets and noise magnitude in real-time according to the model’s convergence state, achieving a balance between privacy protection and model accuracy. Experimental results demonstrate that the proposed scheme achieves high global model accuracy on both the MNIST and CIFAR-10 datasets, effectively resisting up to 40 \(\%\) model poisoning attacks and confirming its security and reliability.