Generation of a driving cycle for battery electric vehicles：A case study of Beijing

Accepted Manuscript Generation of a driving cycle for battery electric vehicles：A case study of Beijing

Huiming Gong, Yuan Zou, Qingkai Yang, Jie Fan, Fengchun Sun PII:

S0360-5442(18)30320-7

DOI:

10.1016/j.energy.2018.02.092

Reference:

EGY 12392

To appear in:

Energy

Received Date:

22 June 2017

Revised Date:

08 February 2018

Accepted Date:

18 February 2018

Please cite this article as: Huiming Gong, Yuan Zou, Qingkai Yang, Jie Fan, Fengchun Sun, Generation of a driving cycle for battery electric vehicles：A case study of Beijing, Energy (2018), doi: 10.1016/j.energy.2018.02.092

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Generation of a driving cycle for battery electric vehicles：A case study of Beijing 1

1

Huiming Gong2 , Yuan Zou , Qingkai Yang , Jie Fan1, Fengchun Sun1 1. Beijing Electric Vehicle Collaboration and Innovation Center, National Engineering Laboratory for Electric Vehicles, Beijing Institute of Technology, Beijing 100081, China 2. National Lab of Auto Performance and Emissions Test, Beijing Institute of Technology, Beijing 100081, China Corresponding Author: Yuan Zou, Tel: 86-10-68944115; fax: 86-10-68944115; e-mail: [email protected]; No. 5 Zhongguancun South Street, Haidian District, Beijing 100081, China.

Abstract Driving cycle, which is widely adopted as a standard measurement procedure for evaluation of fuel economy, emission and driving range, can facilitate vehicle design and performance evaluation of emerging vehicular technologies. However, all current popular driving cycles are developed based on the operation characteristics of conventional vehicles, while the driving characteristics of battery electric vehicles could be quite different, which means traditional driving cycles may not be suitable for evaluating and improving battery electric vehicles. Thus it is important to develop a new driving cycle, which is consistent with real-world situations of battery electric vehicles for the development of new energy auto industry. In this study, the real-world operation data of battery electric vehicles in Beijing are collected with high frequency and the usage and driving characteristics of battery electric vehicles are analyzed based on the real-world data and compared with several standard cycles, such as New European Driving Cycle, Federal Test Procedure-75, and Japan 10-15. Then the Beijing driving cycle is developed using statistic and Markov chain method. The following evaluation proves the new developed driving cycle represents the real-world driving well, which establishes a solid foundation for accurate performance evaluation of battery electric vehicles at least in Beijing. Keyword: battery electric vehicle; driving cycle; energy consumption

1

ACCEPTED MANUSCRIPT

1 Introduction With the rapid development of economy, the number of vehicles in China is rising steadily. According to statistics from MPS (Ministry of Public Security), the vehicle stock number has exceeded 200 million by March 2017 [1]. Growing vehicle fleets have placed a huge burden on oil supply in China. In 2015, China’s automotive fuel consumption accounted for 33% of the total oil consumption, and the proportion is expected to rise to 57% by 2020 [2]. In addition, according to CPEA (China Petroleum Enterprise Association) and COGC (China Oil and Gas Center), the oil import dependence rate reached 65.4% in 2016 [3]. As reported in Ref. [4], vehicles contributed 32% of the total particulate matter with a diameter of 2.5 micrometers or less (PM2.5) in Beijing. As a result, energy security together with serious air pollution has become big concerns in China. Compared with conventional fuel vehicles, New energy vehicles (NEVs), including battery electric vehicles (BEVs), plug-in hybrid vehicles (PHEVs), and fuel cell vehicles (FCVs), have higher energy efficiency and lower exhaust emissions [5,6]. Therefore, in order to improve energy security and air quality, the Chinese government promotes the development of NEVs actively. In 2012, the State Council published “Development Plan of Energy-Saving and New Energy Vehicles Industry” to guide the development of NEVs in China [7]. In 2013, the government of China issued “Notice about promoting the deployment of NEVs with large volume” and decided to continue the generous financial subsidies [8]. In 2015, the State Council published “Made in China 2025” [9], which further strengthened that China would continue to promote the development of NEVs with the goal to master relevant key technologies and achieve the international advanced level. With strong policy support and continuous technology development, the NEV industry in China is rapidly developing with BEV as the priority. Up to March 2017, Chinese domestic auto manufacturers have launched at least 29 battery electric car models, as shown in Table 1. According to statistics from China Association of Automobile Manufacturers (CAAM), in 2016, the sale of BEVs in China reached 409,000 (including 257,000 battery electric cars), which accounted for 81% of the total annual NEV sales, and continued as the top leading NEV market in the world [10]. Table 1 Main battery electric car models of domestic brands in China Brand Model BAIC BJEV

EV150, EV160, EV200, EU260, EH300, EC180, EX260

JAC

iEV4, iEv5, iEV6, iEV6E, iEV7

ROEWE

e50

CHERY

eQ1, eQ, QQ3EV

BYD

E6, E5 300, Song EV300, Qin EV300 2

ACCEPTED MANUSCRIPT GEELY

DiHao EV

CHANGAN

EADO EV, BenBen EV

ZOTYE

E200, Yun 100s, ZhiMa e30

ZD

D1, D2, D2S

As a megacity of China, the car population in Beijing has reached 5.6 million by January 2015 [11], which produces substantial exhaust emission and aggravates the air pollution. To alleviate traffic congestion and environmental pollution, Beijing municipal government has adopted series of policies to limit the purchase as well as usage of cars, such as the vehicle license plate lottery system to control the annual new car sales and traffic restriction based on the number of license plate (every car will be restricted for use one day per five working days). In addition, Beijing has released several policies to stimulate the deployment of BEVs, such as exempting battery electric cars from traffic restriction, which results into an increase of license plates for battery electric cars from 20,000 in 2014 to 60,000 in 2017. Furthermore, Beijing municipal government provides local matching subsidies to NEV buyers. Incentivized by these policies, the number of NEVs in Beijing is increasing remarkably. According to the data from Beijing Municipal Environmental Protection Bureau, the number of BEVs in Beijing has reached 67,000 in September 2016, ranked first in China [12]. In the context of large-scale deployment of electric vehicles in Beijing, studies related to driving cycle are necessary as it plays a vital role in evaluation and improvement of vehicle design. Driving cycle is a velocity-time profile that can represent the real-world driving conditions [13]. There are two types of driving cycles, transient cycles, whose velocity and acceleration are changing at all times, such as FTP75 (Federal Test Procedure-75), and modal cycles, which are composed of a series of constant-velocity or constant-acceleration fragments, such as NEDC (New European Driving Cycle) [14]. Driving cycles can be divided into two categories in terms of application range: standard and non-standard. Standard driving cycles, such as FTP-75 and NEDC, are used by the government to test the fuel/energy consumption of automobiles. Non-standard driving cycles are mainly used in researches from vehicle design to life cycle analysis. For example, Hung, W. T. et al. [15] developed the Hong Kong driving cycle for more accurate vehicle emissions testing and estimation. EstevesBooth, A. et al. [16] generated the Edinburgh cycle to facilitate vehicular assessment. Driving cycle has important applications in fuel/energy consumption evaluation and design procedure of vehicles. Tzirakis, E et al. [17] examined the effects of the driving patterns on fuel consumption and exhaust emissions from cars by developing the Athens Driving Cycle, which was generated from the real-world Athens traffic data. Sze-Hwee Ho et al. [18] developed a representative driving cycle for passenger cars in Singapore to generate more accurate fuel consumption and emissions ratings for various uses (for example, inventory of vehicular emissions and fuel economy labelling). Zhou Bing et al. [19] optimized the transmission ratio of electric vehicles based on the driving cycle derived from the operation data of a front-wheel drive car. A typical driving cycle is supposed to reflect driving conditions exactly so as to provide convincing results in measurement procedure for certification and evaluation 3

ACCEPTED MANUSCRIPT of fuel economy, emissions and newly emerging vehicular technologies. However, due to the differences in road conditions, traffic regulations and driving habits, driving conditions in different regions may have vastly distinct characteristics. Ericsson E et al. [20] carried out an experimental study to compare driving patterns between and within different street-types, drivers and traffic conditions. They found very significant differences between street type and driver, and these factors had a significant impact on all the parameters employed. Seers P et al. [21] developed two different driving cycles of utility vehicles, which indicated that intensive driving is extremely different from any already existing driving cycles. Fellah, M et al. [22] found that in different cycles such as UDDS, LA 92, US 06 and real-world cycle, EV distance greatly varies depending on cycles’ aggressiveness. Tate E et al. [23] compared the power-train characteristics of different electric vehicle types (HEVs, PHEVs and E-REVs) and concluded that driving style varies even in the same region if car types are different. At present, NEDC is regarded as the standard test cycle for the Chinese government. However, due to the distinct driving conditions between China and Europe, NEDC may not reflect the driving style of Chinese drivers suitably, which implies test results based on NEDC may not be consistent with reality. Shaojun Zhang et al. [24] pointed out that a new cycle for the type-approval test for LDPVs (light-duty passenger vehicles) with more real-world driving features is of great necessity in China. Hewu Wang et al. [25] confirmed that using NEDC as Chinese standard test cycle underestimates vehicle fuel consumption and fuel reduction associated with EVs, which may obstruct the promotion of electric vehicles in China. So far, researches about driving patterns in Beijing based on real-world traffic data are not very common. Yuan Zou et al. [26] collected the operational data of electrictaxis in Beijing and found that the average daily driving distance of electric taxis in Beijing is 117.98km, and there are two peaks for the distribution of both departure time and arrival time, which coincide with the rush hours in the morning and evening. Hewu Wang [27] discussed driving range and patterns of private passenger vehicles in Beijing from the sample of 106 cars, 1652 days and 3920 trips collected in the second half of 2012 and early 2013. However, these papers didn’t further generate driving cycles based on the driving pattern analysis, which is the main research contents of this paper specifically. The remainder of the paper is organized as follows. The data acquisition procedure is introduced in Section 2. Section 3 analyses the usage and driving characteristics of EVs in Beijing as well as introduces the generation procedure of the Beijing EV driving cycle. Section 4 concludes the discoveries and main work of this paper.

2 Data collection In this section, information about the 10 selected electric vehicle models in this study is demonstrated and the data collection procedure is described in detail. The data acquisition scheme is firstly using Controller Area Network (CAN) bus to capture the operation data of the electric vehicles, secondly recording the data through the data collect device, and finally translating the raw data from the collecting device into useful 4

ACCEPTED MANUSCRIPT messages with the help of communication protocols. Considering data quality and information integrity, data of five selected vehicles are ultimately chosen to accomplish the study in this paper. 2.1 Vehicle model choice As the EVs demonstration benchmarking city in China, almost all electric vehicle models of all brands can be found in Beijing. The mainstream electric vehicles’ brands in Beijing are BAIC, BYD, JAC, DENZA, CHANGAN, and GEELY. In this paper, 10 electric vehicle models were preliminarily selected as samples, which are shown in Table 2. All of the sample vehicles in this paper are private passenger electric vehicles. The volunteer recruitment notice was released in network forum and EVs club, and data usage agreements were signed with the volunteers. Except for installing the data collect device, no restrictions are imposed on the volunteers, which means the usage and driving behaviors of the sample EVs can be regarded as total autonomous. Table 2 Models and numbers of sample vehicles Model Brand

Number

EV160

BAIC

5

EV200

BAIC

5

EU260

BAIC

6

E50

ROEWE

5

EQ

CHERY

5

iEV4

JAC

5

iEV5

JAC

5

E6

BYD

2

DENZA

DENZA

5

EV150

BAIC

7

2.2 Data acquisition scheme The operation data of electric vehicles can be captured from CAN bus, which is the main communication channel between different ECUs(Electric Control Unit). The data record device used in this study is shown in Figure 1. The device can record the messages on the CAN bus with a high frequency (10Hz). Moreover, it is equipped with GPS (Global Position System). The operation and GPS data of the sample vehicles are recorded in unit of the trip. The detailed technical parameters of the device are shown in Table 3.

5

ACCEPTED MANUSCRIPT

Figure 1 The data record device Table 3 Parameters of the data record device Items Parameters Size

138mm * 100mm * 34mm

Weight

415g

Working voltage

8-32V DC

Communication rate

50kbps, 250kbps, 500kbps, 1000kbps

Power

<3W

GPS precision

2.5m

Communication bus

High rate CAN bus (ISO11898-2)

Storage capacity

32G

Working condition

-40~+80℃, 10-80% RH

Protection level

IP54

After configuration, the data record device is installed on the sample vehicle. Specifically, there are two ways of connecting the device to the CAN bus, as shown in Figure 2. One is through On-Board Diagnostic(OBD) port. Like conventional vehicles, some electric vehicles’ OBD port has reserved pins to communicate with CAN bus. But unfortunately, OBD standards have not been universally adopted on EVs in China yet, and some car types have no OBD pins for connection to the CAN bus. For such models, the CAN bus is tapped within an ECU, such as BMS.

6

ACCEPTED MANUSCRIPT

Figure 2 Two ways of connecting the data record device to CAN bus. (a) Through OBD port. (b) Insert into BMS. The data recorded in the device are raw CAN messages, which need to be translated into readable data using the communication protocol. Due to different communication protocols used for different vehicle types, the data items after decoding are not the same among the sample vehicles. A list of relatively abundant data items(which is the case for sample vehicles from BAIC), by category, is shown in Table 4. Figure 3 shows the velocity, battery current, and battery voltage data of a single trip, and Figure 4 is the corresponding GPS trail. The trip started at Liangxiang, Fangshan, a suburban district of Beijing, and ended at Zhongguancun, Haidian, a downtown area of Beijing. From the velocity profile shown in Fig 3(a), more frequent extended periods of 0 or very low speed appear from suburban to downtown, which suggests the traffic condition is getting congested. Table 4 Quantities recorded by data record device. GPS Battery Motor

Driving information

Longitude

Total current

Motor current

Velocity

Latitude

Total voltage

Motor voltage

Gear

Altitude

SOC

Speed of revolution

Mileage

Time

Remaining capacity

Torque

Acceleration pedal

Temperature

Deceleration pedal

Date Velocity (GPS) Driving direction

7

Vehicle speed (km/h)

ACCEPTED MANUSCRIPT

80 60 40 20 0

0

500

1000

1500 2000 Time(s)

2500

3000

2500

3000

2500

3000

Battery current (A)

(a) 100 50 0 -50 0

500

1000

1500 2000 Time(s)

Battery voltage (V)

(b) 470 460 450 0

500

1000

1500 2000 Time(s)

(c) Figure 3 Example of a single-trip information (a) Vehicle speed (b) Battery current (c) Battery voltage

8

ACCEPTED MANUSCRIPT

Figure 4 The GPS trail of the trip 2.3 Data choice The data record devices are installed on volunteer EVs in sequence, and the data are collected at regular intervals. However, considering data quality and information integrity, not all the data collected are suitable for subsequent analysis. For example, because of different communication protocols used, the data of EQ, iEV4 and iEV5 lack the information about motor current and voltage, while the motor current data of E6 and DENZA are always zeros. Therefore, these vehicle types cannot be used for energy consumption calculation in section 3 and are excluded from the data choice. For the remaining vehicle types, although their data items are satisfied for requirements, the data quality and time span of some sample vehicles are undesirable. Take EV200 as an example, the time spans of the five sample vehicles are two 5 months, two 2 months and one 2 weeks. The sample vehicle of the shortest data time span is not suitable for the following driving cycle generation because it cannot reflect the driver’s driving style comprehensively in only 14 days. In addition, the two sample vehicles with 2-month time span have severe data discontinuity problems, which may be caused by unreliable wire connections. Figure 5 gives an example. The most obvious signs of discontinuity are the three periods with zero values in battery voltage as battery voltage should never become zero during vehicle operation. Correspondingly, discontinuity phenomenon can also be observed in vehicle speed and battery current. Therefore, these two vehicles are not appropriate for following analysis as well and only the remaining two sample vehicles with relatively longer time span and better data quality are selected for subsequent study.

9

Vehicle speed (km/h)

ACCEPTED MANUSCRIPT

20 10 0

0

200

400

600 800 Time(s)

1000

1200

1400

1000

1200

1400

1000

1200

1400

Battery current (A)

（a） 60 40 20 0 -20

0

200

400

600 800 Time(s)

Battery voltage (V)

（b） 300 200

discontinuity

100 0

0

200

400

600 800 Time(s)

（c）

Figure 5 An example of data discontinuity (a) Vehicle speed (b) Battery current (c) Battery voltage Finally, data of five qualified sample vehicles from the database are selected for analysis, as shown in Table 5 (Because two sample vehicles of EV200 are chosen, they are labeled as EV200-1 and EV200-2 to mark the difference). The total number of trips is 3885, and the total mileage is 45384km. Most trips of these five sample vehicles can cover all the representative periods, such as peak and off-peak periods in working days and the traffic peak in holidays, which implies trips used to analyzed in this paper is of abundant representativeness for EVs in Beijing. Table 5 The sample data used in this paper Model Time span Trips Mileage EV160

Twelve months

1417

11313km

EV200-1

Five months

228

14417km

EV200-2

Five months

578

5360km

EU260

Ten months

720

2344km

E50

Twelve months

942

11950km

10

ACCEPTED MANUSCRIPT

3 Analysis and Results In this section, the usage and driving characteristics of EVs in Beijing are analyzed and compared with the standard cycles like NEDC, FTP-75 and JP10-15 (Japan 10-15). The results show that the real-world diving characteristics of EVs in Beijing are quite different from NEDC, the standard driving cycle used by the Chinese government. The Beijing electric vehicle driving cycle is then generated using statistics and Markov process method, and its representativeness is evaluated. 3.1 Usage characteristics of EVs in Beijing Since the sample vehicles in this study are all private EVs, the usage and driving behaviors can be regarded as natural. The definition of the single-trip in this paper is: 1) the mileage is greater than 500m; 2) the positive velocity time proportion is larger than 80%. (1) Distribution of single-trip driving mileage The distribution of single-trip driving mileage is shown in Figure 6, in which the bin size is 2 km. As the result shows, driving mileage is mainly distributed in short range. The mean driving mileage of a single-trip is 14.1km. 47% of the trips are less than 10km, and the trips over 40km merely account for 3.2%. Peter Weldon et al. [28] examined the trip making behavior of a fleet of EV users in Ireland. The comparison of singletrip driving mileage between Beijing and Ireland is shown in Figure 7, where the bin size is 5 km. Compared with Ireland, the EVs’ driving mileage in Beijing is distributed in a relatively longer range.

Figure 6 Distribution of single-trip driving mileage

11

ACCEPTED MANUSCRIPT

Proportion (%)

60 Beijing

50

Ireland

40 30 20 10 0 (0,5] (5,10](10,15](15,20]Single-trip (20,25](25,30] (30,35](35,40] (40,45] (45,50](50,55](55,60](60,80] >80 driving mileage (km)

Figure 7 Comparison of the single-trip driving mileage distribution between Beijing and Ireland. (2) Distribution of single-trip travel time Figure 8 is the distribution of single-trip travel time. It can be seen that the travel time is mainly distributed in short period. The mean travel time is 38 minutes. 80% of the trips are less than 1 hour, and the trips over 2 hours only account for 2%. The short mileage and travel time of EVs in Beijing coincident with the fact that the private EVs are mainly used for commute.

Figure 8 Distribution of single-trip travel time 3.2 Driving characteristics of EVs in Beijing For understanding the characteristics of driving cycles in depth, this paper defines some variables to characterize a driving cycle, as shown in Table 6. For the real-world data, the characteristic variables of every single-trip are calculated, and the mean value of each variable is regarded as the characteristic variable value of the trips made by EVs in Beijing. For standard driving cycles, their characteristic variable values are also 12

ACCEPTED MANUSCRIPT calculated for comparison with the real-world data. Table 6 Characteristic variables of a driving cycle Variables Meaning

Unit

𝑃𝑎

Proportion of acceleration

%

𝑃𝑑

Proportion of deceleration

%

𝑃𝑐

Proportion of cruise

%

𝑃𝑖

Proportion of idle

%

V0 ‒ 10

Proportion of time in speed interval 0~0km/h

%

V10 ‒ 20

Proportion of time in speed interval 10~20km/h

%

V20 ‒ 30

Proportion of time in speed interval 20~30km/h

%

V30 ‒ 40

Proportion of time in speed interval 30~40km/h

%

V40 ‒ 50

Proportion of time in speed interval 40~50km/h

%

V50 ‒ 60

Proportion of time in speed interval 50~60km/h

%

V60 ‒ 70

Proportion of time in speed interval 60~70km/h

%

V70

Proportion of time in speed interval over 70km/h

% 2

%

2

%

𝐴 ‒ 2.5~ ‒ 1.5

2

Proportion of time in acceleration interval -2.5~-1.5 m/𝑠

%

𝐴 ‒ 1.5~ ‒ 0.5

2

%

2

Proportion of time in acceleration interval -0.5~0.5 m/𝑠

%

2

%

2

%

2

%

𝐴 ‒ 3.5 𝐴 ‒ 3.5~ ‒ 2.5

𝐴 ‒ 0.5~0.5 𝐴0.5~1.5 𝐴1.5~2.5 𝐴2.5~3.5 𝐴3.5

Proportion of time in acceleration interval under -3.5m/𝑠 Proportion of time in acceleration interval -3.5~-2.5 m/𝑠 Proportion of time in acceleration interval -1.5~-0.5m/𝑠 Proportion of time in acceleration interval 0.5~1.5 m/𝑠 Proportion of time in acceleration interval 1.5~2.5 m/𝑠 Proportion of time in acceleration interval 2.5~3.5 m/𝑠

2

Proportion of time in acceleration interval over 3.5 m/𝑠 Mean velocity

%

𝑉𝑚 𝑉𝑚𝑝

Mean positive velocity

km/h

𝑉𝑠𝑡𝑑

Standard deviation of velocity

km/h

+

Mean positive acceleration

m/𝑠

‒

Mean negative acceleration

m/𝑠

Standard deviation of acceleration

m/𝑠

𝐴𝑚 𝐴𝑚 𝐴𝑠𝑡𝑑

km/h

2

2

2

(1) Driving characteristics A part of characteristic variables of the real-world data, NEDC, FTP-75 and JP10-15 are shown in Table 7. The mean velocity and mean positive velocity of the sample EVs are 23.96km/h and 29.16km/h respectively, reflecting the congested traffic condition in Beijing. Overall, there are significant differences between the real-world data and existing standard driving cycles. For example, the mean velocity, mean positive velocity, mean positive acceleration and mean negative acceleration of NEDC are respectively 37.85%, 50.82%, 21.95% and 78.57% higher than that of the real-world data, which means the driving style of NEDC is more aggressive compared with the 13

ACCEPTED MANUSCRIPT reality in Beijing. Table 7 A part of characteristic variables of real-world data, NEDC, FTP-75 and JP1015. Variables Real-world data NEDC FTP-75 JP10-15 Vm(km/h)

23.96

33.03

33.88

26.12

Vmp(km/h)

29.16

43.98

41.92

34.41

Vstd(km/h)

18.16

30.77

25.52

22.11

+

2

0.41

0.5

0.49

0.49

‒

2

-0.42

-0.75

-0.53

-0.54

0.45

0.42

0.61

0.45

A m (m/s ) A m (m/s ) 2

Astd(m/s )

(2) Distribution of driving states In this paper, four driving states are defined according to the following criteria: 1) Acceleration: the vehicle acceleration is greater than 0.15m/s2, while the vehicle velocity is positive; 2) Deceleration: the vehicle acceleration is less than -0.15m/s2, while the vehicle velocity is positive; 3) Cruise: the vehicle acceleration is between -0.15m/s2 and 0.15m/s2, while the vehicle velocity is positive; 4) Idle: the vehicle has started, but the velocity is 0. The distribution of driving states is shown in Figure 9. The FTP-75 cycle shows a highly aggressive driving style due to its largest proportion in acceleration and deceleration. The proportion of the four driving states of JP10-15 cycle are approximately equal. Compared with real-world data, the idle and cruise proportion of NEDC are larger, and the acceleration and deceleration proportion are smaller.

14

ACCEPTED MANUSCRIPT

Figure 9 Distribution of driving states (a) Real-world data (b) NEDC (c) FTP-75 (d) JP10-15 (3) Distribution of speed interval The distribution of speed interval is shown in Figure 10. Obviously, FTP-75 has the largest share in high-speed interval. Compared with NEDC, the real-world data takes a larger proportion in lower speed intervals. In fact, compared with other cycles, the speed interval under 30km/h accounts for the largest proportion in the real-world data.

Figure 10 Distribution of speed interval (a) real-world data (b) NEDC (c) FTP-75 (d) JP10-15 15

ACCEPTED MANUSCRIPT (4) Distribution of acceleration interval The distribution of acceleration is shown in Figure 11. The vast majority of 2

acceleration is distributed in the region of (-1.5, 1.5)m/𝑠 for both real-world data and 2

standard driving cycles. Compared with NEDC, the acceleration region (-0.5, 0.5) m/𝑠 shares larger proportion in the real-world data.

Figure 11 Distribution of acceleration (a) real-world data (b) NEDC (c) FTP-75 (d) JP10-15 (5) Speed acceleration frequency distribution The SAFD (Speed Acceleration Frequency distribution) expresses the amount of time spent in specific speed and acceleration bins [29]. The 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 represents the percentage difference between SAFD of standard driving cycles and real-world driving data [13]. The smaller the 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓, the higher the commonality between the cycle in question and the real-world data. The definition of 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 is:

16

ACCEPTED MANUSCRIPT

𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓

∑(𝑆𝐴𝐹𝐷 (𝑖) ‒ 𝑆𝐴𝐹𝐷 = ∑𝑆𝐴𝐹𝐷 (𝑖) 𝑐𝑦𝑐𝑙𝑒

𝑑𝑎𝑡𝑎(𝑖))

2

2

(1)

𝑑𝑎𝑡𝑎

The 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 between NEDC, FTP-75, JP10-15 and real-world driving data are shown in Table 8. The 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 of NEDC, FTP-75 and JP10-15 are 34.75%, 8.16%, 41.46% respectively, which indicates significant differences exist between the standard driving cycles and real-world data. Moreover, the FTP-75 transient driving cycle is more similar to realistic driving conditions because of its smaller 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 compared with NEDC and JP10-15. Table 8 The 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 of NEDC, FTP-75 and JP10-15 against real-world driving data Cycle

𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓(%)

NEDC

34.75

FTP-75

8.16

JP10-15

41.46

From the analysis of usage and driving characteristics of the real-world data, it is noticeable that the driving conditions in Beijing have its own unique characteristics, and are quite distinct with the standard driving cycles. The NEDC cycle has obvious differences with the real-world driving data of EVs in Beijing in the aspects of speed distribution, acceleration distribution, and other driving characteristic variables. Thus, it is important to develop a driving cycle that can represent the realistic driving condition in Beijing for EV evaluation, as addressed in Ref. [25]. 3.3 Representative variables of driving cycles The representative variables are required for filtering the candidate driving cycles [30]. In Ref. [13], regression analysis was performed to determine the least number of statistically significant parameters that influenced the energy consumption of EVs over a driving cycle. Similar work is done in this section, but it needs to be mentioned that the energy consumption data of EVs used in this paper are realistic, not the simulation data like Ref. [13]. First, 27 explanatory variables of a driving cycle are preliminarily selected, as shown in Table 9. The explanatory variables are divided into four categories: velocity related, acceleration related, driving distance and time, and driving characteristics. Then, the correlation coefficient between every two variables in each category is calculated. If two variables demonstrate high correlation, the one who is less correlated with the 17

ACCEPTED MANUSCRIPT response variable will be deleted from the table. For example, the mean velocity and mean positive velocity show high correlation (Figure 12), but the mean positive velocity demonstrates higher relativity with the specific energy consumption of EVs, so the mean velocity is excluded. After two rounds of variable reduction, 13 explanatory variables are finally adopted for regression analysis, as shown in the rightest column of Table 9. The stepwise regression analysis is performed between the remaining 13 variables and energy consumption, which is calculated directly by the current and voltage of the motor in order to avoid auxiliary interference. To make it comparable between different trips and EV models, energy consumption is defined as: energy consumption =

∫𝑢

𝑒𝑚𝑖𝑒𝑚𝑑𝑡

(2)

𝑚∙𝑑

Where 𝑢𝑒𝑚 and 𝑖𝑒𝑚 represent the voltage and current of the motor respectively, m is the curb weight of the vehicle, and d is the driving distance of a trip. Table 9 The initial 27 explanatory variables and the 13 variables for regression analysis Category

Initial explanatory variables

Explanatory

variables

for

regression analysis Velocity

1.Mean velocity

1.Standard deviation of velocity

related

2.Maximum velocity

2.Mean cruise velocity

variables

3.Standard deviation of velocity 4.Mean positive velocity

Acceleration

5.Mean positive acceleration

3.Mean positive acceleration

related

6.Mean negative acceleration

4.Mean negative acceleration

variables

7.Positive acceleration time

5.Standard deviation of positive acceleration

8.Negative acceleration time

6.Standard deviation of negative acceleration

9.Maximum acceleration

7.Percentage

of

time

negative acceleration 10.Minimum acceleration 11.Standard deviation of acceleration 12.Standard deviation of positive acceleration 13.Standard deviation of negative acceleration 14.Percentage

of

time

under

positive

of

time

under

negative

acceleration 15.Percentage acceleration Driving distance

16.Driving distance and

17.Driving time 18

under

ACCEPTED MANUSCRIPT time

18.Total time

Driving

19.Idle time

8.Percentage of idle time

characteristics

20.Percentage of idle time

9.Maximum specific power

21.Number of stops

10.Minimum specific power

22.Number of stops per km

11.Number of stops per km

23.Mean specific power

12.Cruise time

24.Maximum specific power

13.Percentage of cruise time

25.Minimum specific power 26.Cruise time 27.Percentage of cruise time

Figure 12 Correlation between mean velocity and mean positive velocity Finally, 11 variables(including constant term) remain in the regression model, as shown in Table 10. The 11 variables are regarded as the representative variables for filtering the candidate driving cycles. The representative variables of the developed driving cycle are supposed to match the mean value of the representative variables of the real-world data within ± 10%. The comparison of the real-world energy consumption versus the predicted energy consumption by regression model is shown in Figure 13, in which each point represents a trip. It can be seen that nearly all the points concentrate around the straight line with slope one, indicating the predicted energy consumption is very close to the reality. It verifies the chosen representative variables do have a significant influence on the energy consumption of EVs. Table 10 The variables remained in the regression model Variables Coefficients

p-value

Constant

-0.0215

0.000

Standard deviation of velocity

-0.0007

0.000

Mean positive acceleration

0.2896

0.066

Mean negative acceleration

-0.0557

0.000

Standard deviation of positive acceleration

-0.0726

0.000

Percentage of time under positive acceleration

0.2464

0.000

19

ACCEPTED MANUSCRIPT Percentage of time under negative acceleration

-0.1869

0.000

Percentage of idle time

0.0464

0.000

Number of stops per km

0.0008

0.002

Maximum specific power

0.0005

0.000

Minimum specific power

-0.0001

0.000

Figure 13 Comparison of the real-world energy consumption versus the predicted energy consumption by regression model 3.4 Generation of the driving cycle Markov chain has been proved suitable to deal with the random property of driving cycles [31]. If the future states of a system depend only on the current states and are irrelevant with past states, then the system has Markov property. The velocity and acceleration are selected as state variables because the vehicle dynamics can be fully represented by them [24]. The TPM (Transition Probability Matrix) of the state variables is calculated using the real-world driving data of EVs in Beijing, as shown in Figure 14. The TPM consists of the sub-TPMs at different velocity and acceleration states. From the sub-TPM of the current state, the possible state and its probability of the next time could be derived. For 2

example, if the current state is 64km/h and 0.8m/𝑠 , possible velocity and acceleration of the next time and their probabilities can be obtained from the corresponding subTPM, as shown in Figure 14 (b).

20

ACCEPTED MANUSCRIPT

Figure 14 Structure of TPM The procedure of generating a driving cycle is shown in Figure 15. Two points needed to be explained here. First, the next time state is chosen according to the probabilities in the sub-TPM. Second, the termination condition of the driving cycle generation procedure is: 1) the mileage of the cycle is longer than 14.1km, which is the mean mileage of the real-world trip; and 2) the velocity of the last point is zero.

Figure 15 The procedure of generating a driving cycle using Markov chain and statistics method. 3.5 Representativeness evaluation of the BJEV driving cycle The generated BJEV cycle (Beijing Electric Vehicles driving cycle) is shown in Figure 16. The representative variables of BJEV cycle and real-world data are shown in Table 11. Errors of the representative variables between BJEV cycle and the real21

ACCEPTED MANUSCRIPT world data meet the requirement ( ± 10%) mentioned before.

Figure 16 The generated Beijing EVs driving cycle. Table 11 The representative variables of BJEV cycle and real-world data Representative variables Real-world BJEV Relative error data cycle Standard deviation of velocity

18.3812

19.2782

4.88%

Mean positive acceleration

0.3978

0.3675

7.62%

Mean negative acceleration

-0.4136

-0.3961

4.23%

Standard deviation of positive

0.3428

0.3107

9.36%

0.3397

0.3356

1.21%

0.3268

0.3115

4.68%

Percentage of idle time

0.2696

0.2441

9.46%

Number of stops per km

1.3095

1.2418

5.17%

Maximum specific power

31.8958

31.2500

2.02%

Minimum specific power

-44.8345

-43.5185

2.94%

acceleration Percentage of time under positive acceleration Percentage of time under negative acceleration

Tae-Kyung Lee et al. [30] pointed out that if the energy consumption of the generated driving cycle located at the median of the real-world data, then the generated driving cycle represented the real-world driving data well. The distribution of the real-world energy consumption is shown in Figure 17. The energy consumption of BJEV cycle is calculated using the regression model in 3.3, and it locates around the middle of all realworld energy consumption data. Table 12 shows the 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 between real-world data and BJEV, NEDC, FTP-75, JP10-15. Compared with the standard driving cycles, the generated BJEV cycle has the smallest 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 with the real-world data, which means 22

ACCEPTED MANUSCRIPT BJEV cycle is more representative of the real-world EVs driving conditions in Beijing.

Figure 17 Distribution of the real-world energy consumption Table 12 The 𝑆𝐴𝐹𝐷𝑑𝑖𝑓𝑓 between real-world data and BJEV, NEDC, FTP-75 and JP10-15. Cycles

𝑺𝑨𝑭𝑫𝒅𝒊𝒇𝒇 (%)

BJEV

3.91

NEDC

34.75

FTP-75

8.16

JP10-15

41.46

4 Conclusions Driving cycles are the foundations of design and evaluation for electric vehicles. In this paper, volunteers were recruited and the operation data of private passenger EVs in Beijing were collected with high frequency (10Hz). The collected data mainly include the information of power battery, motor, GPS, and vehicle states. The usage and driving characteristics of EVs in Beijing are investigated based on the real-world data. And the driving characteristics of EVs and existing standard cycles are compared. The results show there are significant differences between the real-world data and standard driving cycles in many aspects. Thus it is important to develop and use real-world driving cycles in specific regions to evaluate EVs. The BJEV cycle is generated using statistic and Markov chain method based on the real-world driving data of EVs in Beijing. The results of evaluation show the developed driving cycle represents the real-world data well. The developed driving cycle can assist evaluation, life cycle analysis of EVs in urban areas like Beijing. Analysis results using 23

ACCEPTED MANUSCRIPT the developed driving cycle can be more accordant with the real-world conditions.

Acknowledgments The authors highly appreciate Energy Foundation Grant (G-1512-24144), National Natural Science Foundation China (Grant 51775039 and Grant 51375044) and National Talent Introduction 111 Project (B12022) for supporting this study.

References [1] Ministry of Transport of the People’s Republic of China. The country's vehicle ownership exceeded 300 million. http://www.moc.gov.cn/guowuyuanxinxi/201704/t20170418_2192115.html. Ministry of Transport of the People’s Republic of China, Beijing, P. R. China. [2] Prospective economist. In the future, more than half of our oil will be used in the automotive industry. http://mt.sohu.com/20150928/n422255594.shtml. Prospective economist, Shenzhen, P. R. China. [3] China Petroleum Information Center. China's dependence on foreign crude oil rose to 65.4%. http://news.cnpc.com.cn/system/2017/03/22/001640098.shtml. China Petroleum Information Center, Beijing, P. R. China. [4] Ministry of Environmental Protection of the People’s Republic of China. China motor vehicle environmental management annual report 2016. [5] Teixeira A C R, Sodré J R. Simulation of the impacts on carbon dioxide emissions from replacement of a conventional Brazilian taxi fleet by electric vehicles[J]. Energy, 2016, 115(3):1617-1622. [6] Wang, H., Zhang, X., & Ouyang, M. (2015). Energy consumption of electric vehicles based on real-world driving patterns: a case study of Beijing ☆. Applied Energy, 157, 710-719. [7] The State Council. Energy saving and development plan of new energy automobile industry. http://www.gov.cn/zwgk/2012-07/09/content_2179032.htm. The State Council, Beijing, P. R. China. [8] The Central People's Government of the People's Republic of China. Circular on continuing the popularization and application of new energy vehicles. http://www.gov.cn/zwgk/2013-09/17/content_2490108.htm. The Central People's Government of the People's Republic of China, Beijing, P. R. China. [9] The State Council. Made in China 2025. http://www.gov.cn/zhengce/content/2015-05/19/content_9784.htm. The State Council, Beijing, P. R. China. [10]China Association of Automobile Industry. Economic operation of automobile industry in 2016. http://www.caam.org.cn/xiehuidongtai/20170112/1505203997.html. China 24

ACCEPTED MANUSCRIPT Association of Automobile Industry, Beijing, P. R. China. [11]Beijing Traffic Management Bureau. In 2020, the proportion of green travel in central city was 75%. http://www.bjjtgl.gov.cn/jgj/mtlj/446948/index.html. Beijing Traffic Management Bureau, Beijing, P. R. China. [12]Beijing Municipal Environmental Protection Bureau. Beijing has reached 67 thousand pure electric vehicles. http://env.people.com.cn/n1/2016/0913/c1010-28711436.html. Beijing Municipal Environmental Protection Bureau, Beijing, P. R. China. [13]Brady, J., & O’Mahony, M. (2016). Development of a driving cycle to evaluate the energy economy of electric vehicles in urban areas. Applied Energy, 177, 165-178. [14]Jiang Ping. (2011). Investigation of the Driving Cycle Construction of Mixed Roads in the City. (Doctoral dissertation, Hefei University of Technology). [15]Hung, W. T., Tong, H. Y., Lee, C. P., Ha, K., & Pao, L. Y. (2007). Development of a practical driving cycle construction methodology: a case study in Hong Kong. Transportation Research Part D Transport & Environment, 12(2), 115-128. [16]Esteves-Booth, A., Muneer, T., Kirby, H., Kubie, J., & Hunter, J. (2001). The measurement of vehicular driving cycle within the city of Edinburgh. Transportation Research Part D, 6(3), 209-220. [17]Tzirakis, E., & Pitsas, K. (2006). Vehicle emissions and driving cycles: comparison of the Athens driving cycle (adc) with ece-15 and European driving cycle (edc). Global Nest, 8(3), 282-290. [18]Ho, S. H., Wong, Y. D., & Chang, W. C. (2014). Developing Singapore driving cycle for passenger cars to estimate fuel consumption and vehicular emissions. Atmospheric Environment, 97, 353-362. [19]Zhou, B., Jiang, Q., Yang, Y., & Wang, J. (2010). Analysis of energy consumption and powertrain parameters optimization of BEV based on running cycle. IEEE, International Conference on Computer-Aided Industrial Design & Conceptual Design (Vol.2, pp.1284-1290). IEEE. [20]Ericsson, E. (2000). Variability in urban driving patterns. Transportation Research Part D, 5(5), 337-354. [21]Seers, P., Nachin, G., & Glaus, M. (2015). Development of two driving cycles for utility vehicles. Transportation Research Part D Transport & Environment, 41, 377-385. [22]Fellah, M., Singh, G., Rousseau, A., Pagerit, S., Nam, E., & Hoffman, G. (2009). Impact of Real-World Drive Cycles on PHEV Battery Requirements. SAE World Congress & Exhibition. [23]Tate, E., Harpster, M. O., & Savagian, P. J. (2009). The electrification of the automobile: from conventional hybrid, to plug-in hybrids, to extended-range electric vehicles. SAE International Journal of Passenger Cars - Electronic and Electrical Systems, 1(1), 156-166. [24]Zhang S, Wu Y, Liu H, et al. Real-world fuel consumption and CO2, (carbon dioxide) emissions by driving conditions for light-duty passenger vehicles in 25

ACCEPTED MANUSCRIPT China[J]. Energy, 2014, 69(5):247-257. [25]Wang, H., Zhang, X., & Ouyang, M. (2015). Energy consumption of electric vehicles based on real-world driving patterns: a case study of Beijing ☆. Applied Energy, 157, 710-719. [26]Zou, Y., Wei, S., Sun, F., Hu, X., & Shiao, Y. (2016). Large-scale deployment of electric taxis in Beijing: a real-world analysis. Energy, 100, 25-39. [27]Wang H, Wu L, Hou C, et al. A GPS-based research on driving range and patterns of private passenger vehicle in Beijing[C]// Electric Vehicle Symposium and Exhibition. IEEE, 2014:1-7. [28]Weldon, P., Morrissey, P., Brady, J., & O’Mahony, M. (2016). An investigation into usage patterns of electric vehicles in Ireland. Transportation Research Part D Transport & Environment, 43, 207-225. [29]Dubarry, M., Svoboda, V., Hwu, R., & Liaw, B. Y. (2010). A roadmap to understand battery performance in electric and hybrid vehicle operation. Journal of Power Sources, 174(2), 366-372. [30]Lee, T. K., & Filipi, Z. S. (2010). Synthesis and validation of representative realworld driving cycles for Plug-In Hybrid vehicles. Vehicle Power and Propulsion Conference (pp.1-6). IEEE Xplore. [31]Gong, Q., Midlam-Mohler, S., Marano, V., & Rizzoni, G. (2011). An iterative markov chain approach for generating vehicle driving cycles. Sae International Journal of Engines (1), 1035-1045.

26

ACCEPTED MANUSCRIPT Title: Generation of a driving cycle for electric vehicles：A case study of Beijing

Highlights:  Operational data of fifty private electric vehicles in Beijing were collected.  The usage and driving characteristics of Beijing electric vehicles were analyzed.  Significant differences exist between real-world data and standard driving cycles.  The Beijing driving cycle was developed using statistic and Markov chain method.