基于联邦深度学习的工业网络物理系统入侵检测数据集
收藏资源简介：
在复杂和动态化的网络安全威胁背景下，内生安全的研究成为解决网络安全问题的重要方向。内生安全强调系统具备自我免疫、自我适应和自我恢复能力，通过主动感知和抵御潜在攻击，实现从被动防护向主动防护的转变。本文提出了一种基于联邦深度学习的内生网络安全数据集，旨在弥补现有数据集在动态场景、多节点协作及隐私保护等方面的不足。数据集融合了复杂网络环境、多种攻击行为及隐私保护机制，不仅包含正常通信和已知攻击流量，还模拟了零日攻击以评估系统应对未知威胁的能力。数据采集过程基于高仿真的虚拟环境，覆盖多种工业协议和典型攻击行为，如数据篡改、流量阻断和命令注入。数据处理通过标准化、标签生成及时间序列重构确保质量，并提供多维度特征支持联邦学习模型的训练与验证。实验结果表明，数据集可用于评估内生安全系统在主动防护、自我恢复等方面的能力，为相关研究提供可靠的测试基准和分析工具。
Against the backdrop of complex and dynamic cybersecurity threats, research on endogenous security has emerged as a critical direction for addressing cybersecurity issues. Endogenous security emphasizes that a system should possess self-immunity, self-adaptation, and self-recovery capabilities, and achieve the transition from passive defense to active defense through proactive perception and defense against potential attacks. This paper proposes an endogenous cybersecurity dataset based on federated deep learning, aiming to compensate for the shortcomings of existing datasets in dynamic scenarios, multi-node collaboration, and privacy protection. The dataset integrates complex network environments, diverse attack behaviors, and privacy protection mechanisms. It not only includes normal communication traffic and known attack traffic but also simulates zero-day attacks to evaluate the system's ability to cope with unknown threats. The data collection process is based on a highly simulated virtual environment, covering multiple industrial protocols and typical attack behaviors such as data tampering, traffic blocking, and command injection. Data processing ensures quality through standardization, label generation, and time series reconstruction, and provides multi-dimensional features to support the training and verification of federated learning models. Experimental results show that this dataset can be used to evaluate the capabilities of endogenous security systems in active defense, self-recovery, and other aspects, providing a reliable test benchmark and analysis tool for relevant research.
