On the Feasibility of Proﬁling Electric Vehicles through Charging Data Ankit Gangwaly Aakash Jainy and Mauro Contix

2025-04-24 0 0 682.77KB 9 页 10玖币

侵权投诉

On the Feasibility of Proﬁling Electric Vehicles

through Charging Data

Ankit Gangwal†, Aakash Jain†, and Mauro Conti§

†International Institute of Information Technology, Hyderabad

§University of Padua

Email: gangwal@iiit.ac.in, aakash.jain@students.iiit.ac.in, mauro.conti@unipd.it

Abstract—Electric vehicles (EVs) represent the long-term

green substitute for traditional fuel-based vehicles. To encourage

EV adoption, the trust of the end-users must be assured.

In this work, we focus on a recently emerging privacy threat

of proﬁling and identifying EVs via the analog electrical data

exchanged during the EV charging process. The core focus of

our work is to investigate the feasibility of such a threat at scale.

To this end, we ﬁrst propose an improved EV proﬁling approach

that outperforms the state-of-the-art EV proﬁling techniques.

Next, we exhaustively evaluate the performance of our improved

approach to proﬁle EVs in real-world settings. In our evaluations,

we conduct a series of experiments including 25032 charging

sessions from 530 real EVs, sub-sampled datasets with different

data distributions, etc. Our results show that even with our

improved approach, proﬁling and individually identifying the

growing number of EVs appear extremely difﬁcult in practice;

at least with the analog charging data utilized throughout the

literature. We believe that our ﬁndings from this work will further

foster the trust of potential users in the EV ecosystem, and

consequently, encourage EV adoption.

I. INTRODUCTION

The growing concerns related to the climate crisis have

led a global movement to adopt green and renewable energy

for a sustainable future. Electric Vehicles (EVs) represent a

long-term ecological substitute for fossil fuel-based vehicles.

EVs are even perceived as the key patrons for achieving

near zero carbon footprint [26, 27]. Today, EVs are becoming

increasingly popular as well as gaining widespread adoption.

As a representative example, the global sales of EVs in Q1 ’21

were over 2.5 times of their sales in Q1 ’20 [1]. As an

estimate [12], the annual EV sales will reach over 31.1 million

by 2030; which will represent approximately 32% of new car

sales worldwide. Furthermore, some vehicle manufactures have

plans to produce only EVs by 2040 [2].

With the increasing adoption of EVs, the demand for

their charging apparatus, i.e., Electric Vehicle Supply Equip-

ments (EVSEs), is naturally increasing. Although EV charg-

ing equipment can be installed on residential premises, their

absence in public spaces is often seen as a limiting factor,

This is an extended version of our paper in VehicleSec 2023 (co-located

with NDSS Symposium 2023), San Diego, CA.

which restricts EV users to not travel far away from the

charging station. Various governments, as well as industry

players, are working to solve this issue by increasing the

presence of EVSEs in public spaces. For instance, the USA,

Germany, and China have allocated dedicated funds to develop

the EV charging network [12] in their countries. On the

other hand, companies are installing EVSEs in their parking

lots for their employees [26]. Thus, we can expect major

growth in publicly available EVSEs in the coming years that

will reduce infrastructure availability concerns, increase users’

convenience, and may further boost EV adoption.

Unlike the refueling process of conventional vehicles, the

charging process of EVs involves complex communication

protocols and information exchange between users/EVs and

EVSEs infrastructure. To initiate a charging session on a

public EVSE, a user has to book a charging session, negotiate

power requirements, authorize the session and payment for the

service, and ﬁnally station the vehicle for the duration of the

charging process. The overall charging process of EVs can

be divided into two parts: (i) resource negotiation phase and

(ii) actual charging phase [17, 20].

As the interactions (between the user and EVSE infras-

tructure) in the former phase involve exchanging private infor-

mation, such interactions are protected by the state-of-the-art

communication protocols and cryptographic mechanisms [3,

14]. The interactions (between EV and EVSE infrastructure)

in the latter phase primarily focus on transferring energy

to recharge the vehicle and do not involve sharing of any

personal information. Therefore, the signals in the charging

phase are neither authenticated nor coded; these signals are

exchanged in the clear. Consequently, an attacker may exploit

such unprotected signals as a side channel to gain information

about the EV, e.g., its battery behavior [23].

Motivation: As the majority of public EVSEs are installed

without proper physical access control or supervision, such

equipments are accessible to anyone [4, 8]. Thus, attackers

targeting EVSE infrastructure can modify [3] EVSE’s physical

port and gather data related to the charging phase of benign

users’ EVs. In fact, recent works [6, 7, 23] demonstrate how to

use such data/signals to proﬁle EVs with certain assumptions.

Such attacks, if possible in real-world settings, can severely

threaten users’ privacy because attacker(s) - who have access

to multiple public EVSEs - can track the movements of users

who use compromised charging stations. In this paper, we

investigate the extent and feasibility of such proﬁling of EVs

in real-world scenarios. One of the major beneﬁts of such an

investigation is that it will help the community to understand

the actual magnitude of EV proﬁling threat.

Symposium on Vehicles Security and Privacy (VehicleSec) 2023

27 February 2023, San Diego, CA, USA

ISBN 1-891562-88-6

https://dx.doi.org/10.14722/vehiclesec.2023.23021

www.ndss-symposium.org

arXiv:2210.05433v2 [cs.CR] 14 Feb 2023

The key idea behind EV proﬁling is that each EV exhibits

unique physical characteristics during a charging session. More

precisely, when the State of Charge (SoC) of the battery

goes above a certain threshold (say, over 60% or 80%), the

current and voltage drawn by the vehicle solely depend on the

battery’s implementation. Therefore, these physical properties

- which can differ from one EV to another - can be used to

create signatures of EV batteries; consequently, the signature

of EVs. Authors in work [23] demonstrate modeling the

behavior of EV batteries from their charging data. Their work

extracts features from analog charging signals and uses that

information for battery proﬁling via clustering-based approach.

EVScout attack (originally EVScout1.0 [7], and recently EVS-

cout2.0 [6]) further improved such proﬁling of EVs by utilizing

different machine learning techniques.

Contributions: In this paper, we begin with improving the

state-of-the-art of EV proﬁling. To understand the impact of

the improved EV proﬁling approach at scale in the real world,

we emphasize on the multi-class classiﬁcation (contrary to

binary classiﬁcation considered in the state-of-the-art proﬁling

approach) to evaluate its efﬁcacy in proﬁling/identifying a

particular EV. Furthermore, we consider datasets that vary in

size, balancing, and distribution to closely simulate different

settings. The major contributions of this paper are as follows:

1) We propose an improved EV proﬁling approach that

outperforms the state-of-the-art, i.e., EVScout.

2) We exhaustively evaluate the quality of our improved

approach at scale by considering a signiﬁcantly large

dataset of charging sessions from real EVs as well as

different classiﬁcation techniques, etc.

Organization: The remainder of this paper is organized as

follows. Section II presents a brief summary of the funda-

mental concepts related to our work. Section III explains our

threat model and attack infrastructure. Section IV elucidates

the implementation details of our approach. Section V reports

our experimental evaluations. Section VI comments on the

limitations of the current practices to proﬁle EVs. Section VII

concludes the paper.

II. BACKGROUND

The concept of using electric or analog data for the

purpose of user proﬁling has been extensively studied in the

literature [11]. The central aspect of the EV charging system is

the EVSE infrastructure. A central control unit is responsible

for monitoring the operation of all EVSEs connected to a par-

ticular grid. These operations include appropriate scheduling

of charging processes (keeping track of power availability and

maximum allowed load for the network, etc.) and constituting a

gateway for secure communication between the grid and an EV

(to allow user authentication, etc.). It is important to note that

EVSEs are typically part of a complex network, where they can

communicate with each other, an EV, or the control unit via

appropriate communication interfaces. Such communications

happen over a secure channel that can be wireless or wired.

An EV user must be connected to the control center via a

car or mobile application. The security considerations of this

communication network is addressed by strong cryptographic

tools and mechanisms [13].

The physical port on EVSEs that connects it to an EV is

built upon SAE J1772 Standard [25] (cf. Fig. 1). According

to this standard, a port consists of ﬁve lead connectors. Out

of these ﬁve leads, three are are connected to the grid via

relays while the other two leads are used for signaling. In

particular, these two leads individually carry proximity signal

and pilot signal. The proximity signal veriﬁes whether the

physical connection between the EV and EVSE’s port is safe

and that the communication or charging can proceed. On the

other hand, the pilot signal serves as a communication medium

between the EV and EVSE to signal charging level, etc.

The charging characteristics of the battery units used in

EVs also play a part in the proﬁling process. Most battery

units deployed in EVs today are lithium-ion batteries [9].

The charging process for standard lithium-ion batteries is

distinctive, where the drawn current and voltage follow a

ﬁxed proﬁle [20]. In particular, its charging process can be

of two types, i.e., Constant Power/Constant Voltage (CP/CV)

and Constant Current/Constant Voltage (CC/CV). In this work,

we only consider the latter as sufﬁcient data is not publicly

available for CP/CV charging-based EVs. The CC/CV charg-

ing method consists of two phases:

1) Constant Current: It is the primary phase of charging,

during which the current passed remains constant while

the voltage across the battery terminals varies.

2) Constant Voltage: It is the latter phase of charging, during

which the current passed drops while the voltage across

the battery terminals remains constant.

The transition from the CC to CV phase is roughly preset,

but it is also ascribed by the state and condition of the

EV’s battery. Such transition threshold varies between 60%

and 80% of the battery’s SoC. Similar to the state-of-the-

art, our approach utilizes analog signal data (e.g., current

and pilot signals) obtained from the CC/CV charging phases

for EV proﬁling. Nonetheless, our work differs in various

aspects, including an improved proﬁling algorithm, modeling,

classiﬁcation approach, etc.

III. THREAT MODEL

EV proﬁling attacks (e.g., EVScout [6, 7]) present in the

literature assume that an attacker is capable of installing a

physical device - typically over EVSEs’ physical port - to

intercept the analog signals exchanged between EVSEs and

EVs. With such a device in place, the attacker(s) can intercept,

record, or transmit the observed signals to the attacker(s),

where they can process the collected signals. It is worth

mentioning that if such a device has wireless transmission

capabilities, then tracing the original attacker(s) can become

even more difﬁcult. By tampering multiple EVSEs, the at-

tacker(s) can have access to multiple charging sessions of

different (often, even the same) EVs. Therefore, the attacker(s)

can exploit such charging data to proﬁle the unique charging

behavior of an EV’s battery; which essentially means the

proﬁle of that EV.

The data obtained by such a data collection practice will be

unlabeled because the extracted signal is analog in nature and

does not contain any personally identifying details. Manual

monitoring, utilizing cameras, or collusion with local staff

can make the attack sophisticated. Nevertheless, by gathering

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

OntheFeasibilityofProlingElectricVehiclesthroughChargingDataAnkitGangwaly,AakashJainy,andMauroContixyInternationalInstituteofInformationTechnology,HyderabadxUniversityofPaduaEmail:gangwal@iiit.ac.in,aakash.jain@students.iiit.ac.in,mauro.conti@unipd.itAbstractElectricvehicles(EVs)representthelong-t...

展开>> 收起<<

On the Feasibility of Proﬁling Electric Vehicles through Charging Data Ankit Gangwaly Aakash Jainy and Mauro Contix.pdf

共9页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

On the Feasibility of Proﬁling Electric Vehicles through Charging Data Ankit Gangwaly Aakash Jainy and Mauro Contix

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: