Detection of Cyber Attacks in Electric Vehicles Using a Deep Neural Network

Abstract

The paper focuses on the identification of cyber-attacks in electric cars based on CAN bus signals anomalies detection. One of the objectives of the paper is to show that cyber-attacks may result in taking control over a module in an electric vehicle. On the other hand, the article suggests a protection method based on anomalies’ detection using a deep neural network that can be used for increasing the level of CAN bus transmission security. This paper discusses an experiment that was conducted using an additional intruder module on CAN bus operating in bridge mode, which task is to change specific control signals in a way that is unnoticeable to the vehicle’s supervisory system. The aim of the research is to analyse the time dependencies between the control signals transmitted on the CAN bus in a situation where a cyber attack on the vehicle has been carried out. By using deep neural learning algorithms, it is possible to detect anomalies suggesting the existence of a cyber-attack and to take countermeasures. It should be emphasised that the presented methodology for detecting anomalies of control signals on CAN buses can also serve as a tool of protection against taking control over the vehicle.