Electric Vehicles CAN Bus Cyber-Attacks Detection Using Machine Learning and Deep Learning Techniques

Abstract

The number of Electric Vehicles (EVs) users have increased drastically in the recent years. Although this increase in the number of EVs in replacement of the conventional fossil fuel based vehicles can be very beneficial to the environment, great concerns about the cybersecurity of these EVs and how vulnerable they are to cyber-attacks have risen. The EVs Controller Area Network (CAN) bus protocol which is the protocol used in EVs to send messages and communicate between the different Electronic Control Units (ECUs) in the EV, does not adopt any authentication, authorization, or accounting methods that can provide cybersecurity for the bus and thus can serve as a target for attackers to exploit and damage the EVs. The smart grid-transport nexus includes two major elements, which are the transport network elements, including EVs and their charging stations (EVCSs), and the smart grid network elements, including the generation nodes and the demand nodes. The aim of this thesis research is to develop and propose an EV CAN bus cyber-attacks detection techniques based on the Adaptive Neuro Fuzzy Inference System (ANFIS) algorithm in addition to implementing different machine learning and deep learning techniques, and test their performances in detecting and classifying the cyber-attacks. The first three tasks of this thesis were a parametric analysis that modelled and simulated the voltage and frequency stability effects of demand side cyber-attacks on different nexus configurations. The nexus configurations integrated synchronous generators, solar PV systems, and induction motor loads. The results have demonstrated that demand side cyber-attacks can cause complete system failures and blackouts and that the integration of induction motor loads into the nexus can improve the nexus resilience against demand side cyber-attacks and can act as a passive protection technique against them. The fourth task of this thesis proposed and discussed a new and unique demand side cyber-attack vector against the nexus through the EVs CAN bus that can impact the nexus stability. The discussed attacks that exploit the unique attack vector and impact the nexus stability are the Denial of Service (DOS) and impersonation attacks. The fourth task also proposed and implemented new CAN bus cyber-attacks detection techniques based on the combination of Adaptive Neuro Fuzzy Inference System (ANFIS) algorithm with Subtractive Clsutering (SC) and Fuzzy C-Means clustering (FCM). The proposed techniques are ANFIS-SC and ANFIS-FCM, where their detection performances were compared with the different machine learning and deep learning based techniques using different statistical metrics. The results showed that the proposed ANFIS-SC and ANFIS-FCM detection techniques had excellent performance and were successful in the detection of CAN bus cyberattacks. The proposed ANFIS-SC and ANFIS-FCM detection techniques had higher testing accuracies and overall F1 scores and thus had better performance compared to standard machine learning based techniques and the state of the art deep learning based detection techniques. Moreover, the proposed ANFIS-SC and ANFIS-FCM detection techniques had decent testing time compared to the standard machine learning based techniques and had the lowest testing time compared to the state of the art deep learning based detection techniques.

Date of Award	Aug 2023
Original language	American English
Supervisor	KHALED AL JAAFARI (Supervisor)