Priv-IDS: A Privacy Protection and Intrusion Detection Framework for In-Vehicle Network

Abstract

Intelligent connected vehicle (ICV) is equipped with advanced on-board sensors, controllers, actuators and other equipment of the new generation of vehicles, integrated with modern communication and network technology, to achieve intelligent information exchange and sharing. As an international standardized communication protocol, controller area network (CAN) plays an important role in vehicle communication. However, due to the CAN is plaintext broadcast communication, lack of encryption technology, CAN faces the challenge of malicious attack and privacy disclosure. In this paper, a machine learning method Priv-IDS based on local differential privacy (LDP) is proposed to protect the privacy of CAN data and detect malicious intrusion. The method performs random perturbation on CAN data and detects malicious attacks through temporal convolutional network (TCN). We propose a $\alpha \beta$-LDP method, which ensures data availability as much as possible while protecting data privacy. This method provides a way to solve the problem of privacy disclosure caused by CAN data in machine learning intrusion detection. Based on the standard data set, this scheme is compared with other vehicle intrusion detection methods. The experimental results show that the proposed intrusion detection method is not different from other intrusion detection methods in terms of accuracy and time efficiency, but it is the first intrusion detection method to better protect CAN data based on LDP.