Scholarlink

<p>Over the past decade, Software-Defined Networking (SDN)-based Industrial Cyber-Physical Systems (ICPS) have been developed at a large scale and have become more vulnerable to extensive, complex cyberattacks attributed to their degree of interconnectedness and dependence on centralized control systems. Meanwhile, the IDS based on deep learning approaches have achieved a promising accuracy but still suffer from severe limitations in ICPS, like vulnerability to slowly evolving attacking patterns and dependence on traditional feature engineering techniques that may ignore subtle but effective signs of malicious behaviour. To address these gaps, we present a framework called EGOA-DL-IDS that uses Explainable AI and the Grasshopper Optimization Algorithm (GOA), combined with a Hybrid Adaptive Loss Function (HALF), to provide resistance to adversarial cases and enable robust and interpretable feature selection on intrusion detection datasets. Together with a Bi-GRU classifier, which is optimized via Bayesian optimization, the model obtains a detection rate of 98.99% while maintaining interpretability and robustness. The potential contribution of the proposed EGOA-DL-IDS model includes improving the security of SDN-based industrial cyber-physical systems. This proposed model addresses the identification of cyberattacks in addition to the risks brought about by centralized control, prevents the loss of critical infrastructure, and strengthens the resilience of enterprises in terms of industrial operations.</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

EGOA -DL-IDS: explainable intrusion detection in SDN-based industrial cyber physical system with optimized deep learning techniques

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Sivamohan S,
  • Sasi Rekha Sankar,
  • Krishnaveni S,
  • B. Jothi,
  • Thiyagu T

摘要

Over the past decade, Software-Defined Networking (SDN)-based Industrial Cyber-Physical Systems (ICPS) have been developed at a large scale and have become more vulnerable to extensive, complex cyberattacks attributed to their degree of interconnectedness and dependence on centralized control systems. Meanwhile, the IDS based on deep learning approaches have achieved a promising accuracy but still suffer from severe limitations in ICPS, like vulnerability to slowly evolving attacking patterns and dependence on traditional feature engineering techniques that may ignore subtle but effective signs of malicious behaviour. To address these gaps, we present a framework called EGOA-DL-IDS that uses Explainable AI and the Grasshopper Optimization Algorithm (GOA), combined with a Hybrid Adaptive Loss Function (HALF), to provide resistance to adversarial cases and enable robust and interpretable feature selection on intrusion detection datasets. Together with a Bi-GRU classifier, which is optimized via Bayesian optimization, the model obtains a detection rate of 98.99% while maintaining interpretability and robustness. The potential contribution of the proposed EGOA-DL-IDS model includes improving the security of SDN-based industrial cyber-physical systems. This proposed model addresses the identification of cyberattacks in addition to the risks brought about by centralized control, prevents the loss of critical infrastructure, and strengthens the resilience of enterprises in terms of industrial operations.