基于证据深度学习的CAN网络入侵检测框架

Intrusion Detection Framework for CAN Networks Based on Evidence Deep Learning

随着移动通信技术在智能自动驾驶系统中的持续发展,保障车载通信数据的安全已成为交通系统安全的一个重要环节,面对黑客可能通过CAN总线网络远程操控车辆的威胁,现有网络框架虽能检测已知攻击,但在识别未知攻击时表现不佳。为此,本研究提出一种融合证据深度学习的检测框架,该框架由数据预处理模块、数据分析模块和攻击检测模块组成。预处理模块通过独立热编码技术,以提升数据质量和适应性;数据分析模块通过生成对抗网络(GAN)技术增强该框架的泛化能力并模拟攻击场景;攻击检测模块应用了证据深度学习,提高了框架在应对未知攻击时的不确定性处理能力。该框架在开源汽车黑客数据集和基于奇瑞EXEED RX车型自主构建的数据集上进行了测试。实验结果表明,该框架在检测未知攻击时,相比于传统的基于softmax的分类网络综合性能提高了24.5%。

With the continuous development of mobile communication technologies in intelligent autono⁃mous driving systems,securing vehicular communication data has become pivotal for transportation safety.Faced with threats of hackers remotely manipulating vehicles through the CAN bus network,existing frameworks can de⁃tect known attacks but falter in identifying location-based attacks.A detection framework integrating evidence-based deep learning is proposed in this paper,comprising data preprocessing,analysis,and attack detection modules.The preprocessing module employs independent hot encoding to enhance data quality and adaptability.The analysis module utilizes Generative Adversarial Networks(GANs)to bolster the framework's generalization and simulate at⁃tack scenarios.The attack detection module harnesses evidence-based deep learning to enhance the framework's ca⁃pability in handling uncertainties from unknown attacks.The framework is tested on an open-source car hacking data⁃set and a dataset constructed based on the Chery EXEED RX model.The test results show that the framework im⁃proves the overall performance by 24.5%in detecting unknown attacks compared to traditional classification proba⁃bility-based networks.