汽车信息安全:面向总线网络的伪造攻击检测技术  

Automotive Cyber-security:Detection Technique of Masquerade Attacks for the Bus Network

作　　者：魏洪乾 时培成[4] 张幽彤[1,2] WEI Hongqian;SHI Peicheng;ZHANG Youtong(School of Mechanical Engineering,Beijing Institute of Technology,Beijing 100081;Key Laboratory of Low Emission Vehicles in Beijing,Beijing 100081;Vehicle Measurement,Control and Safety Key Laboratory of Sichuan Province,Chengdu 610039;School of Mechanical Engineering,Anhui Polytechnic University,Wuhu 241000)

机构地区：[1]北京理工大学机械与车辆学院,北京100081 [2]清洁车辆北京市重点实验室,北京100081 [3]汽车测控与安全四川省重点实验室,成都610039 [4]安徽工程大学机械工程学院,芜湖241000

出　　处：《机械工程学报》2024年第10期476-486,共11页Journal of Mechanical Engineering

基　　金：国家重点研发计划(2021YFB3101500);国家自然科学基金(52202461);中国博士后自然科学基金(2022TQ0032,2022M710380);汽车新技术安徽省工程技术研究中心开放基金(QCKJ202202A);汽车测控与安全四川省重点实验室开放基金(QCCK2023-001)资助项目。

摘　　要：当前的智能网联汽车正面临着潜在的信息安全挑战。比如,汽车CAN总线(Controller aera network,CAN)采用明文方式传输消息,缺少发送源电子控制单元(Electronic control unit,ECU)的身份认证和信息加密机制。因此,如何定位异常报文的发送源对于保证网联汽车的信息安全具有重要的研究意义。基于此,提出基于总线信号特征的ECU身份识别技术用于定位报文发送源,并检测ECU伪造攻击:首先根据CAN总线的电平信号提取关键的身份特征参数,包括边沿跳变时间、平台时间、高电平电压众数等;然后,利用轻量化的Softmax分类器对提取的身份特征进行离线训练并建立在线的学习模型。实车测试结果表明,与传统方法相比,提出的方法能够提高将近10%的ECU识别精度,而且该方法可以有效地检测到潜在的ECU伪造攻击和报文篡改攻击等。此外,进一步评估ECU工作温度对相关特征参数的影响,间接地验证了所提方法的强鲁棒特性。综上,提出的方法有效地解决了传统CAN网络缺乏身份认证的缺陷,保证了智能网联汽车的信息安全。Intelligent connected vehicles(ICVs)are facing a huge challenge of cyber security.For instance,automotive CAN transmits messages with the plain texts,which lacks of the identity recognition of transmitter electronic control units(ECUs)and encryption mechanism.Therefore,how to identify the transmitter of abnormal messages plays a significant role for the automotive cyber-security.Accordingly,an ECU identification recognition technique for masquerade attacks based on the signal features of CAN bus is proposed.Specifically,the core identity parameters based on voltages of CAN are extracted including the rising-falling edge time,plateau duration and mode of high voltages;then,the lightweight Softmax classifier is utilized to train the characteristic parameters offline and constructs the online learning model.The real-world experiments manifest that compared with the traditional method,the proposed method could improve the ECU identification accuracy by about 10%,which is also effective to detect the masquerade attacks.Besides,effects of the operation temperature on the extracted parameters are also evaluated which has indirectly validates the strong robustness of the proposed method.All in all,the proposed method has addressed the defects of CAN network and guaranteed the cyber-security of ICVs.

关 键 词：智能网联汽车 信息安全 总线网络 电子控制单元 身份识别 

分 类 号：TG156[金属学及工艺—热处理]