Conference PaperPDF Available

Defining the accuracy of real-world range estimations of an electric vehicle


Abstract and Figures

Range anxiety is a major barrier for the mass adoption of electric vehicles (EVs), a contributing factor to this is the variability of the predicted range remaining presented to the driver in the vehicle. This study aims to better understand the causes of potential inaccuracies and how ITS can help resolve these issues. Eleven participants completed 141 logged journeys, with results showing that range (as predicted by the EV and presented to the driver) was overestimated by approximately 50% in comparison to journey distance. Driving style had the most significant impact on range prediction accuracy, where a more aggressive driving style led to greater inaccuracies. However, journey distance and type of road driven, which can be calculated from Satnav systems, were factors which were correlated with having a significant effect on range accuracy. Therefore incorporating these into future range prediction algorithms has the potential to increase the accuracy of information and subsequently increase driver trust.
Content may be subject to copyright.
AbstractRange anxiety is a major barrier for the mass
adoption of electric vehicles (EVs), a contributing factor to this
is the variability of the predicted range remaining presented to
the driver in the vehicle. This study aims to better understand
the causes of potential inaccuracies and how ITS can help
resolve these issues. Eleven participants completed 141 logged
journeys, with results showing that range (as predicted by the
EV and presented to the driver) was overestimated by
approximately 50% in comparison to journey distance. Driving
style had the most significant impact on range prediction
accuracy, where a more aggressive driving style led to greater
inaccuracies. However, journey distance and type of road
driven, which can be calculated from Satnav systems, were
factors which were correlated with having a significant effect
on range accuracy. Therefore incorporating these into future
range prediction algorithms has the potential to increase the
accuracy of information and subsequently increase driver trust.
Index TermsElectric vehicle; Driver behaviour; Driver
information systems; Intelligent transportation systems; Range
anxiety; Range accuracy; Real-World driving
Electric Vehicles (EVs) have long been proposed as a
viable method to accelerate the reduction in carbon
emissions of road transport. Carbon emissions in the EU (i.e.
EU-28) have been reducing overall steadily since 1990 and
by 1.3% between 2011 and 2012, with a strong decline in
road transport emissions of 5% in 2012 alone [1]. However,
this reduction may be slightly exaggerated by the global
economic downturn which resulted in a reduction in both
passenger and freight transportation by up to 20% in some
EU countries [1]. Despite this, road transportation in general
has a key role in delivering CO2 reductions in line with the
International Energy Agency’s (IEA) “2DS” (2°C Scenario),
where the sector’s potential share of overall CO2 reductions
world-wide is predicted to be 21% by 2050. In order to meet
this share, three-quarters of all vehicle sales by 2050 would
need to be plug-in electric of some type [2].
Research supported by the Technology Strategy Board through the
WMG centre High Value Manufacturing Catapult
S. A. Birrell is with WMG, University of Warwick, Coventry, CV4 7AL,
UK (+44 24 7657 3752;
A. McGordon is with WMG, University of Warwick, Coventry, CV4
7AL, UK (
P. A. Jennings is with WMG, University of Warwick, Coventry, CV4
7AL, UK (
Research and innovation into electrically and alternatively
powered vehicles is continuing at pace, specifically
surrounding battery technology and light-weighting. Despite
these efforts there still remain many barriers to adoption for
electric vehicles, which have thus far limited EV sales
primarily to early adopters of innovative technologies and
company fleets looking to reduce carbon emissions. A report
for the European Council suggested that the main barriers
for adoption were: high purchase price; range anxiety;
uncertainties associated with battery life; and other factors
relating to new and unfamiliar technology [3]. Range anxiety
is a common term given when drivers experience anxiety
about their cars ability to cover the distance required before
needing to be recharged. It occurs almost exclusively in EV
drivers because of limited charging infrastructure [4]. More
complex than the phenomenon of range anxiety is how to
combat it in users in order to remove one of the barriers to
mass uptake of EVs. Long term goals would be to increase
electric range to over 300 miles (similar to that of a
traditional internal combustion engine (ICE) vehicle), and /
or to make plug-in charging facilities as commonplace as
fuel stations are today. Both of these are vastly expensive
options in terms of research and infrastructure costs, and are
unlikely to be achieved in the next 10 years. Intelligent
Transportation Systems (ITS) can offer solutions to these
problems in the shorter term by developing integrated
charging management and services (such as pre-booking and
payment of charging points), improved traffic management
and delivering eco-driving advice [5]. In addition by
improving route guidance to minimise energy consumption
of EVs, by taking into account congestion levels and
topography [5, 6]. However accurate range prediction, which
is subsequently presented to the diver in the vehicle,
underpins these solutions.
Range prediction in EVs is not a simple procedure with
algorithms needed which take into account aspects such as
Peukert’s coefficient, temperature and depth of discharge
[7]. The driver also has significant influence on range
predictions with the use of heating, ventilation and air
conditioning (HVAC) and auxiliary use, or driving style.
Road parameters also have an important impact on energy
usage for a journey (and hence range prediction) these
include topography, number of stop/starts and sharp turns,
traffic congestion and speed limit. Many of the parameters
mentioned above are not unique to EV range prediction but
common to ICE distance to empty calculations too. However
Defining the Accuracy of Real-World Range
Estimations of an Electric Vehicle
Stewart A. Birrell, Andrew McGordon, and Paul A. Jennings
this coupled with the reduced absolute range (300-500 miles
for an ICE compared to approximately 100 with an EV) and
lack of convenient refuelling leads to range anxiety on
behalf of EV drivers. For these reasons research has shown
that EV users deploy ‘safety buffers’ for ensuring that they
do not run out of range, these buffers have been shown to be
as much as 25% of vehicle range capacity [8, 9].
One of the key factors for increasing anxiety is a lack of
trust in the range prediction presented via the in-vehicle
display and the subsequent use of high safety buffers. The
accuracy of range prediction has also been identified as one
of the top five priorities for research and development into
ITS to support electromobility [5]. In order to facilitate
improvements in range predictions first we need to
understand where the inaccuracies are present and what
impact the have on EV range. This paper evaluates range
predictions over 140 journeys taken by 11 drivers on
familiar commuting routes during real-world driving.
A. Data Collection
A 2011, right-hand drive, Nissan Leaf full battery EV was
used for the data collection. According to the owner’s
manual [10] this specific EV has a fully charged range of
‘approximately 100 miles’, adding ‘the majority of drivers
will experience vehicle ranges between 62 - 138 miles based
on the many factors that affect vehicle range’. No
modifications were made to the vehicle; however a GPS data
logger was installed for all trials.
Eleven participants took part in the trials; with journeys
taken predominately over familiar routes (i.e. commute to
and from work). Each driver typically completed six
journeys in a one week period, with a minimum of four
journeys being considered the threshold for inclusion in the
analysis. The mean age of the participants was 38 years
(Standard Deviation = 9.3 years), all were employees of
WMG at the University of Warwick, with positions from
technicians, engineers and researchers. Given the nature of
the trials those who volunteered were intrinsically interested
and curious regarding EVs and new technology; however
were novice EV drivers.
Two phases of data collection were adopted with four of
the 11 participants completing trials in both phases 1 and 2.
The two phases were adopted for two reasons, firstly to
evaluate any possible effects of range prediction over a
greater array of environmental temperature (phase 1 was UK
summer, phase 2 autumn), and secondly to allow more data
to be collected (table 1).
Phase 1
Phase 2
Mean Temperature
16.2 deg C
13.1 deg C
No of Participants
No of Trials
B. Procedure
All participants were emailed an information sheet
containing details of the study, they were also given some
basic information about the EV (i.e. charging, conditions of
use etc.). Also to avoid participants being placed in difficult
situations regarding the distance of journeys they wish to
take, they were informed of the range prediction accuracy
with wording similar to that contained in the owner’s
manual. Inclusion criteria for the study were simply that they
were registered to drive a University owned vehicle on the
University’s insurance policy.
On the participants first day of the trials they were given
an introduction to the driving controls and in-vehicle
systems of the EV, they were also shown how to plug in and
charge the vehicle. The principal form of data collection for
the study (see below) was data captured from the in-vehicle
systems relating to range and energy use. This data were
collected via a log sheet, which the participants were also
instructed how to complete. After a brief practice drive
participants were left to their own devices with the EV to
experience use in real-world driving scenarios.
C. Variables Collected
As indicated above data were collected via a datasheet
based on what they could read on the in-vehicle information
system (IVIS; figure 1), or what they could select as driving
related options. Route characteristics were also recorded.
Below is listed the data that were collected, and options
given to the participants to select from:
Figure 1. Nissan Leaf in-vehicle driver interface (taken from Nissan Leaf
2011 Owner’s Manual [10]).
Time and Date of journey taken
Temperature of the external environment according to
IVIS (figure 1, number 5)
Driving Mode: The EV has two selectable driving modes,
either ‘D’ or ‘Eco’. D, or Drive mode, is intended for all
normal driving and delivers all of the available power when
accelerating, with regenerative braking delivering similar
deceleration when removing the foot from the throttle as a
typical ICE vehicle. Eco mode is intended to help extend the
driving range by limiting power when accelerating, more
aggressive regenerative braking and reduced HVAC
capability. Depending on external conditions and previous
driving behaviour, when fully charged the Eco mode
predicts that between 10 and 20 more miles could be
completed on a single charge
Start and End of journey predicted range remaining
according the IVIS (figure 1, number 12)
HVAC: Heating and air conditioning use draws a
significant amount of power from the battery and hence
reduces range, therefore recording accurately its use is
essential. Options given for the driver to record were: AC;
Heater; Fan; Occasional (indicating cyclic or intermittent use
HVAC); Demist; or None. More than one option could be
AUX: Again use of the auxiliary power systems will draw
power from the battery, options available were: Radio; CD;
MP3; Satnav; 12v Charger; Lights; Wipers; or None (Status
screen). For analysis this was simplified to: Infotainment;
Navigation; Safety; or None. More than one option could be
Driving Style: In an attempt to establish driver behaviour
a self-rating of driving style was also recorded, options
included: Eco; Conservative; Normal; Progressive;
Aggressive; or Mixed. For the analysis this was simplified
to: Eco; Normal; or Progressive
Traffic flow was also estimated by the driver, options
available were: High; Med; Low; Mixed Density; or Free
flow (i.e. busy but moving freely)
Route: The percentage of driving time (not distance) on
each type of road category was estimated by participants. A
percentage of driving time was recorded for: Urban (town or
city); Rural (or countryside); Inter-Urban (major A-roads
linking conurbations); Motorway (aka freeway, highway
Distance of the journey was also recorded.
Comments: In the final sections participants were
encouraged to make free-form comments on: Range
Estimations; IVIS; Driveability; and Charging
Statistical data analysis was conducted in SPSS 21 and
significance was accepted at p<0.05. The nominal data
(Driving Mode, HVAC, Aux, Driving Style and Traffic
Density) were analysed using a One-Way ANOVA, and the
numerical data (Temperature, Start and End range, Route
and Distance) analysed using regression analysis.
D. Normalisation
A normalisation process was needed in order to evaluate
the accuracy of the EVs ability to predict range used for a
specific journey. Simply using distance minus range used
would not accurately reflect the prediction accuracy for long
or short journeys; therefore the following equation was used:
Normalised Range = (Journey Distance Range Used)
Journey Distance
A normalised range value of zero would indicate that the
range prediction was completely accurate (i.e. range used
equals journey distance). If the equation returned a positive
value then the EV was underestimating the range needed to
complete the journey, a negative value was an
overestimation (i.e. the range used was greater than the
journey distance).
A. General Findings
Results from the study (table 2) show that the average
journey distance completed by the participants in this study
was 16.8 miles (SD = 13.1). The maximum journey distance
was 65.4 miles, with 56.7% of journeys being longer than 10
miles. The average range used was 25.3 miles, and
normalised range accuracy was -0.48. This suggests that the
range predicted for an average journey was almost 50%
greater than the actual journey distance. This results in a
real-world theoretical maximum range for this particular EV
(when fully charged) of approximately 70 miles. This is
similar to results found in the literature which showed
average energy consumption of 1.5% State of Charge
(SoC)/mile which extrapolated to a theoretical range of 66.7
miles (for a vehicle with a 100 mile certified range) [7], or
1.9% SoC/mile with a maximum range of 53 miles (70 mile
certified range) [11].
Results from this current study and surrounding literature
suggest that inaccurate real-world range prediction is a
common EV problem and not specific to any particular
manufacturer. Anecdotal evidence from consumer blogs
suggests that newer EV have improved range accuracy
predictions and increased real-world ranges over first
generation vehicles. This could be as result of more efficient
management of the HVAC system
, by presenting actual
state of charge on the in-vehicle display
, or over-the-air
updates to range algorithms
Phase 1
Phase 2
Twenty-five (or 17.3%) of the 141 trips taken resulted a
positive (or zero in two cases) normalised accuracy (i.e.
journey distance being greater than or equal to range used;
figure 2), and of these 25 just over three-quarters were less
than 10 miles in length. Of the remaining 82.7% of trips
which had a negative normalised accuracy approximately
two-thirds of these were greater than 10 miles in length. This
highlights that the longer the trip, the more problems the EV
had with accurately predicting range. Given that many such
journey prediction systems use history based algorithms, it
could be expected that a longer journey would give the
algorithms more time to adapt to driver style or traffic
conditions, and hence become more accurate. This was not
observed in this current study with increased distance
leading to reduced accuracy.
Figure 2. Normalised range accuracy for each individual trial. Dotted line
indicates zero, or range used equates to journey distance.
Figure 3 shows the majority of trips were conducted in
Drive mode, with no AC, heating or auxiliary load drawing
power from the vehicle. Drivers were also predominantly
driving in a self-rated eco style in low traffic densities, over
a mixed driving route. These findings suggest that drivers
were actively trying to increase the range of the EV by
adopting behaviours and driving in real-world situations
which could be considered advantageous for maximising
energy usage from an EV. This further highlights the
importance of accurate range prediction for users, as shown
in this study range prediction was about 50% out even with
users being conscious of the influence of power demand on
EV range. What was not amended was driving route for
individual journeys, i.e. avoidances of high speed roads or
taking the shortest (as opposed to quickest) routes.
Figure 3. Results relating to a) Drive Mode; b) HVAC; c) Auxiliary usage
(Info means Infotainment, i.e. radio, CD or MP3); d) Driving Style; e)
Traffic Density; and f) Route.
As mentioned previously two phases of data collection
were completed separated by 3-4 months. Statistical
analysis revealed that only ‘Temperature’ and ‘Start Range’
differed significantly (p<0.05) between the phases. Mean
temperature being 16.2 compared to 13.1° C, and start range
being 75.2 and 84.1 miles for phases 1 and 2 respectively.
B. Accuracy Correlations with Dependent Variables
When evaluating the data as a complete set of individual
trials, identifying trends within the data compared to
normalized range accuracy was problematic. The broad
range of data from the individual trials seemed to obscure
potential differences, or simply highlight that making
assumptions on range prediction was very difficult. Of the
variables assessed (as outlined in the methodology) only
‘Driving Style’ showed a statistically significant (F(1,140) =
3.01, p<0.05) effect on normalized range accuracy. Post-hoc
analysis (Bonferroni corrected) revealed that this difference
was significant between Eco and Progressive driving styles
(figure 4); showing that when adopting a self-rated Eco
driving style normalized range accuracy improved to -0.34
from -0.75 for a progressive driving style.
Figure 4. Driving style verses normalised range accuracy. Columns
represent mean data and error bars standard error. Asterisk (*) represents
significant (p<0.05) difference from the Progressive condition.
Auxiliary use showed a trend (p=0.080) for an effect on
range accuracy, with post-hoc analysis suggesting the use of
safety devices (e.g. lights and windscreen wipers) over no
auxiliary use being responsible the decreased prediction
accuracy (p<0.05). Whilst the lights of a vehicle may draw
up to 200 W of power from the batteries, this is not a huge
draw compared to the likes of the HVAC system which
can be up to 3 kW. It is the assumption of the authors that
other driver behavior factors may be accountable for this
difference, such as increased driving speed and reduced
congestion, or reduced visibility leading to increased need
for sharper braking and hence reduced opportunities for
regenerative braking.
External temperature has been shown in previous research
to be strongly linked to battery capacity [12, 13] and hence
range of an EV. However in this current real-world study
figure 5 shows that ambient external temperature recorded
at the start of the journey had no significant effect (p>0.05)
on range accuracy. Reasons for this could be that start range
was already reduced due to the temperature, hence accuracy
was not actually affected. Secondly that the mean
temperature changes from 13.1 16.2° C were not sufficient
to significantly affect battery capacity, with lab tests
typically utilizing a far greater range in exposed
temperatures (e.g. -25 to +45° C) [12].
Figure 5. External temperature verses normalised range accuracy.
C. Analysis Between Similar Journeys
As highlighted thus far in the results, when considering
individual trips as unique data points, identifying notable
trends is very difficult (with the exception of driving style).
To aid the analysis a subset of the data were considered
separately for analysis. This subset included only journeys
taken by an individual participant which were similar in
nature, i.e. the same distance over the same route, typically a
commute to and from work. As previous results showed no
difference between the two phases of data collection, all data
were combined. A minimum of four journeys were needed
in order to be considered in the analysis. This resulted in
data from seven participants, totaling 66 individual journeys
being identified.
Figure 6. Normalised range accuracy against journey distance for all
journeys logged in the study (a), subset of journeys identified as being
similar in nature (b), and average of participants fixed journeys (c).
From this data more representative mean values for the
journeys could be calculated, as all journeys taken by
individual participants were consistent. An example of this
is illustrated in figure 6, where graph (a) shows very little
correlation when all of the data points for normalized range
accuracy are plotted against journey distance (R2 = 0.003).
When we consider each data point for similar journeys only
(graph b), a comparable pattern is observed with a slightly
stronger trend for a decrease in range accuracy as journey
distance increases (R2 = 0.136). However when we use
average range accuracy for each individual participant over
their specific journey (graph c) we see a significant
correlation between distance and normalized accuracy (R2 =
Aspects of the route driven also showed interesting
correlations when considering this subset of data. Figure 7
shows that both the percentage of the journey completed on
either motorway or in the urban environment can have an
impact on range prediction accuracy. An increase in the
percentage of the journey completed on higher speed
motorways had a negative effect on accuracy, whereas
increased urban driving improved average range accuracy.
Despite this, data shows that for only one participant where
urban driving equated to 60% and motorway was 0% of the
total driving time, a positive normalized range accuracy
value was returned (i.e. range used was less then journey
distance). Reasons for these observations may be that an
increase in motorway driving may be indicative of an
increase in journey distance (and vice-versa for urban
driving) which as shown above resulted in a decrease in
range accuracy, or that motorway driving involves constant
high draw from the battery with little opportunity for
regenerative braking, which again is the opposite to urban
Figure 7. Normalized range accuracy verses % of journey completed on
either (a) motorway or (b) urban roads.
D. Subjective Comments on Range Accuracy
Participant’s subjective views were also collected at the
end of their trials, these comments are summarized below:
Participant 1: For a few of my journeys I got to my
destination with more range than I started with, other times
for he some journey it would drop by 20 miles, I have no
idea why as my driving was the same (I think)
Participant 2: Needs better damping; Range increase in
urban and plummets on motorway; 10 miles in town -
mileage estimation went from 93 to 96, battery blocks went
from full to full-2, charge time went to 2 hours, what’s the
difference? Range estimation in D feels more accurate than
E; Approx. 25 miles reduction in first 3 miles, then 10 in
next 24 miles
Participant 3: Inconsistent and confusing; Would prefer a
band rather than a single number; Estimated range seems to
bear little relationship to driving style - only regen activity;
Battery fuel gauge was far more reliable estimate
Participant 4: Fine on short distances, way out on mid-
range distances that have a short section on fast roads; Being
stuck in traffic seems to lower the range estimations on
otherwise similar drives
Participant 6: I thought the range accuracy was quite
unpredictable, steady driving would reduce range estimate
quite dramatically and this can be worrying for 25+ miles
Participant 7: It seems to go down rather quickly which
means you feel conscious of it the whole time
Participant 9: 90→12 on vehicle was actually on 56
A common theme from the subjective comments was the
unpredictable nature of the range estimations, with little
understanding as to what is the actual root cause of the
inaccuracies. One thing identified by the drivers, and
supported by the data analysis, was that range accuracy
decreased with journey distance. What is also very apparent
from the comments is the type of language used, with words
like ‘feel conscious’, ‘unpredictable’, ‘inconsistent and
confusing’, ‘no idea’, ‘dramatically’ and ‘plummets’. These
are very emotive words and are typical of range anxiety
experienced by novice users of EVs. However research has
shown that whilst range anxiety (e.g. being aware of range
while driving) does not reduce with EV experience, the
explicit worry of range anxiety (e.g. being concerned of
running out of energy while driving) does, as users increase
their ability to understand changes in the EV’s
instrumentations [14].
E. Limitations and Future Research
Whilst a large pool of individual journeys were collated
for this study enabling some interesting conclusions to be
made, the findings only describe behaviour for one particular
range estimation algorithm implemented in this specific
vehicle. In addition the authors had no control over the range
algorithm. Future studies will make greater use of the GPS
logged data to better understand objective (rather than self-
reported) driving style, but also to evaluate enhanced GPS
features such as elevation and predicted verses actual road
speeds. Utilising different EVs and an increased number of
drivers will also increase transferability of the results.
Results from this study have shown that during real-world
EV driving trials the accuracy of the range predictions,
which are presented to the driver via the IVIS, are
overestimated by approximately 50% compared to actual
journey distance. This resulted in a real-world theoretical
maximum range for this particular EV of 70 miles. Self-
reported driving style had a statistically significant impact
on range accuracy, with the more aggressively driven
journeys experiencing less accurate range estimations.
However, external temperature was seen to have no effect on
accuracy. When considering a subset of the data, which
consisted of only similar journeys, the journey distance and
percentage completed on motorways led to increased range
prediction inaccuracies, whereas urban driving decreased
these inaccuracies.
With respect to future ITS, in the near-term results from
this study suggest that journey distance and road category
could be incorporated into future enhanced route guidance
systems (in addition to road topography and traffic
congestion) to improve the accuracy of range prediction. In
addition to this EV manufacturers should consider the
grouping of repeatable and predictable journeys for data
analysis, rather than analysis all individual trips, to refine
range algorithms as more accurate assumptions can be made.
In the longer term, further advancements could be delivered
by the self-learning car, as driver behaviour measured over
time can be incorporated to further increase range accuracy
predictions or deliver specifically tailored eco-driving
advice. However what intelligent systems will not be able to
control will be the driver behind the wheel, as the human is
still the biggest factor for increasing the available range of
an EV.
[1] European Environment Agency, “Annual European Union greenhouse
gas inventory 19902012 and inventory report 2014,” EEA Tech. Rep
No 9/2014, Jun. 2014.
[2] International Environment Agency, “Global EV outlook:
Understanding the electric vehicle landscape until 2020,” IEA Tech.
Rep. Apr. 2013.
[3] N. Wallis, B. Lane, “Electric vehicles: Improving consumer
information to encourage adoption”, European Council for Energy
Efficient Economy, Tech. Rep. 4-514-13, 2013.
[4] T. Wellings, J. Binnersley, D. Robertson, T. Khan, “Human machine
interfaces in low carbon vehicles: Market trends and user issues”,
Low Carbon Vehicle Technology Project, Technical Report: HMI 2.0,
Dec. 2010
[5] Y. Hubner, P. T. Blythe, G. A. Hill, M. Neaimeh, C. Higgins, “ITS for
electric vehicle an electromobility roadmap,” IET and ITS
Conference on Road Transport Information and Control (RTIC 2012),
Sept. 2012, pp. 1-5.
[6] M. Neaimeh, G. A Hill, Y. Hubner, P. T. Blythe, “Routing systems to
extend the driving range of electric vehicles, IET Intelligent
Transport Systems, vol. 7, pp. 327-336.
[7] S. Heath, P. Sant, B Allen, Do you feel lucky? Why current range
estimation methods are holding back EV adoption, Hybrid and
Electric Vehicles Conference 2013 (HEVC 2013), 2013, pp. 1-6.
[8] S. Carroll, C.Walsh, M. Burgess, M. Harris, S. Mansbridge, N. King,
L. Bunce, “The ultra-low carbon vehicle demonstrator programme
Final Report: Assessing the viability of EVs in daily life,” Tech. Rep.
Sept. 2013.
[9] T. Franke, J. F. Krems, Interacting with limited mobility resources:
Psychological range levels in electric vehicle use, Transportation
Research Part A: Policy and Practice, vol. 48, Feb. 2013, pp. 109-122
[10] Nissan, “2011 Leaf Owner’s Manual,” available at
[11] C. Walsh, S. Carroll, A. Eastlake, P. Blythe, “Electric vehicle driving
style and duty variation performance study,” Cenex Tech.Rep. Nov.
[12] C. Antaloae, J. Marco, F. Assadian, A Novel Method for the
Parameterization of a Li-Ion Cell Model for EV/HEV Control
Applications,” IEEE Trans. Vehicular Tech., vol. 61, Nov. 2012, pp.
[13] H-S, Song, J-B. Jeong, B-H. Lee, D-H. Shin, B-H. Kim, T-H. Kim, H.
Heol, “Experimental Study on the Effects of Pre-Heating a Battery in
a Low-Temperature Environment,” IEEE Vehicle Power and
Propulsion (VPPC) Conference, Oct. 2012, pp. 1198-1201.
[14] M. Nilsson, “Does range anxiety exist? Experiences from the ELVIRE
project,” 5th International Conference on Applied Human Factors and
Ergonomics (AHFE), Jul. 2014, pp. 1-7.
... Range anxiety can be reduced by off-board charging stations but off-board charging stations are designed only for public and commercial charging stations. The analysis of actual EV trip data is presented in [110]. The range estimation provided by EV proved to be overestimated on more than half of the occasions. ...
Full-text available
Transportation electrification is a pivotal factor in accelerating the transition to sustainable energy. Electric vehicles (EVs) can operate either as loads or distributed power resources in vehicle-to-grid (V2G) or vehicle-to-vehicle (V2V) linkage. This paper reviews the status quo and the implications of transportation electrification in regard to environmental benefits, consumer side impacts, battery technologies, sustainability of batteries, technology trends, utility side impacts, self-driving technologies, and socio-economic benefits. These are crucial subject matters that have not received appropriate research focus in the relevant literature and this review paper aims to explore them. Our findings suggest that transitioning toward cleaner sources of electricity generation should be considered along with transportation electrification. In addition, the lower cost of EV ownership is correlated with higher EV adoption and increased social justice. It is also found that EVs suffer from a higher mile-per-hour charging rate than conventional vehicles, which is an open technological challenge. Literature indicates that electric vehicle penetration will not affect the power grid in short term but charging management is required for higher vehicle penetration in the long-term scenario. The bi-directional power flow in a V2G linkage enhances the efficiency, security, reliability, scalability, and sustainability of the electricity grid. Vehicle-to-Vehicle (V2V) charging/discharging has also been found to be effective to offload the distribution system in presence of high EV loads.
... This algorithm considered different user driving styles as well as electrochemical reactions of lead-acid batteries. Birrell et al.[61] found 50 % overestimation of EV DTE over 141 logged journeys. The authors reported variation of driving styles as the most significant contributor to inaccurate DTE estimations. ...
Full-text available
Electric vehicle (EV) drivers require reliable distance to empty (DTE) indication to plan their trips. In the current study, feed forward neural networks based soft sensors were designed to accurately predict DTE in a Ford Escape EV. The proposed DTE soft sensors were trained on actual drive cycle data and rated DTE using Levenberg Marquardt, Bayesian Regularization and Scaled Conjugate Gradient algorithms. Regression models were also developed for comparisons. Primary results show that the Bayesian Regularization trained soft sensor network with eleven hidden layer neurons achieved the highest testing accuracy (99.64%) among the two layered networks, followed by the Levenberg Marquardt (two layered, eleven hidden layer neurons, testing accuracy 99.62%) and Scaled Conjugate Gradient trained networks (two layered, seven hidden layer neurons, testing accuracy 99.49%). The linear and non linear regression models attained 96.19% and 97.53% accuracies respectively. Deeper soft sensor networks yielded better prediction accuracies at higher computation times. The five layered Bayesian Regularization trained network (with ten neurons in each hidden layer) maximised DTE prediction accuracy to 99.89%, but at the cost of 1175% more training time as compared to the best performing two layered network soft sensor. An optimal choice of prediction accuracy considering reasonable computation timescales can help reduce range anxiety of EV users significantly.
... However, the limited charging infrastructure and battery capacity restrict their usability [3] [4]. Furthermore, the estimation of the current vehicle range lacks accuracy and trustworthiness [5] [6]. The combination of these factors can lead to range anxiety; the idea that the currently available energy might be insufficient to reach the planned destination, thus causing the driver to feel anxious and uncertain. ...
The usability of vehicles is highly dependent on their energy consumption. In particular, one of the main factors hindering the mass adoption of electric (EV), hybrid (HEV), and plug-in hybrid (PHEV) vehicles is range anxiety, which occurs when a driver is uncertain about the availability of energy for a given trip. To tackle this problem, we propose a machine learning approach for modeling the battery energy consumption. By reducing predictive uncertainty, this method can help increase trust in the vehicle's performance and thus boost its usability. Most related work focuses on physical and/or chemical models of the battery that affect the energy consumption. We propose a data-driven approach which relies on real-world datasets including battery related attributes. Our approach showed an improvement in terms of predictive uncertainty as well as in accuracy compared to traditional methods.
... Conversely, Energy Usage and Battery remained consistent or increased significantly after the driving events. Confirmation of vehicle range was important to participants during driving events, and it has been found that the communication of electric vehicle range is important to the development of trust in the technology [74]. Furthermore, it is notable that participants tended to use a visual icon (Battery) alongside more detailed explanations of 'why' (Energy Usage), highlighting the importance of this form of combined information presentation, not only for future automated vehicles, but also for current electric vehicles. ...
Partially automated vehicles present a large range of information to the driver in order to keep them in-the-loop and engaged with monitoring the vehicle's actions. However, existing research shows that this causes cognitive overload and disengagement from the monitoring task. Adaptive Human Machine Interfaces (HMIs) are an emerging technology that might address this problem, by prioritising the information presented. To date, research aiming to define the driver's glance fixation behaviour in a partially automated vehicle to contribute towards an adaptive interface is scarce. This study used a unique three-day longitudinal driving simulator study design to explore which information drivers in a partially automated vehicle require. Twenty-seven participants experienced nine partially automated driving simulations over three consecutive days. Nine information types, developed from standards, previous studies and industry collaboration, were displayed as discrete icons and presented on a surrogate in-vehicle display. Unique to the literature, this study showed that the recorded eye-tracking data demonstrated that usage of the information types changed with longitudinal driving simulator use. This study provides three key contributions: first, the longitudinal study design suggest that single exposure HMI evaluations may be limited in their assessment. Secondly, this study has methodologically shortlisted a list of nine information types that can be used in future studies to represent future partially automated vehicle interfaces. Finally, this is one of the first studies to characterise glance behaviour for partially automated vehicles. With this knowledge, this study contributes important design recommendations for the development of adaptive interfaces.
Full-text available
Under the vehicle-to-grid (V2G) concept, electric vehicles (EVs) can be deployed as loads to absorb excess production or as distributed energy resources to supply part of their stored energy back to the grid. This paper overviews the technologies, technical components and system requirements needed for EV deployment. Electric vehicles as a service (EVaaS) exploits V2G technology to develop a system where suitable EVs within the distribution network are chosen individually or in aggregate to exchange energy with the grid, individual customers or both. The EVaaS framework is introduced, and interactions among EVaaS subsystems such as EV batteries, charging stations, loads and advanced metering infrastructure are studied. The communication infrastructure and processing facilities that enable data and information exchange between EVs and the grid are reviewed. Different strategies for EV charging/discharging and their impact on the distribution grid are reviewed. Several market designs that incentivize energy trading in V2G environments are discussed. The benefits of V2G are studied from the perspectives of ancillary services, supporting of renewables and the environment. The challenges to V2G are studied with respect to battery degradation, energy conversion losses and effects on distribution system.
Full-text available
Partially automated vehicles are increasing in prevalence and enable drivers to hand over physical control of the vehicle’s longitudinal and latitudinal control to the automated system. However, at this partial level of automation, drivers will still be required to continuously monitor the vehicle’s operation and take back control at any time from the system when required. The Society of Automotive Engineers (SAE) defines this as Level 2 automation and consequently a number of design implications arise. To support the driver in the monitoring task, Level 2 vehicles today present a variety of information about sensor readings and operational issues to keep the driver informed; so appropriate action can be taken when required. However, existing research has shown that current Level 2 HMIs increase the cognitive workload, leading to driver cognitive disengagement and hence increasing the risk to safety. However, despite this knowledge, these Level 2 systems are available on the road today and little is known about what information should be presented to drivers inside these systems. Hence, this doctorate aimed to deliver design recommendations on how HMIs can more appropriately support the driver in the use of a partially automated Level 2 (or higher) vehicle system. Four studies were designed and executed for this doctorate. Study 1 aimed to understand the information preferences for drivers in a Level 2 vehicle using semi-structured interviews. Participants were exposed to a 10 minute, Level 2 driving simulation. A total of 25 interviews were conducted for first study. Using thematic analysis, two categories of drivers: ‘High Information Preference’ (HIP) and ‘Low Information Preference’ (LIP) were developed. It was evident that the drivers' expectations of the partial automation capability differed, affecting their information preferences and highlighting the challenge of what information should be presented inside these vehicles. Importantly, by defining these differing preferences, HMI designers can be more informed to design effective HMI, regardless of the driver’s predisposition. Building on this, an Ideas Café public engagement event was designed for Study 2; implementing a novel methodology to understand factors of trust in automated vehicles. Qualitative data gathered from the 35 event attendees was analysed using thematic analysis. The results reaffirmed the importance of the information presented in automated vehicles. Based on these first two studies, it was evident that there was an opportunity to develop a more robust understanding of what information is required in a Level 2 vehicle. Information requirements were quantitatively investigated through two eye-tracking studies (Studies 3 and 4). Both used a novel three- or five-day longitudinal study design. A shortlist of nine types of information was developed based on the results from the first two studies, regulatory standards and collaborations with Jaguar Land Rover experts. This was the first shortlist of its kind for automated vehicles. These 9 information types were presented to participants and eye- tracking was used to record their information usage during Level 2 driving. Study 3 involved 17 participants and displayed only steady state scenarios. Study 4 involved 27 participants and introduced handover and warning events. Across both studies, information usage changed significantly, highlighting the methodological importance of longitudinal testing over multiple exposures. Participants increased their usage of information confirming the vehicle’s current state technical competence. In comparison, usage decreased of future state information that could help predict the future actions of the vehicle. By characterising the change in information usage, HMI designers can now ensure important information is designed appropriately. Notably, the ‘Action Explanation’ information, that described what the vehicle was doing and why, was found to be consistently the most used information. To date, this type of information has not been observed on any existing Level 2 HMI. Results from all four studies was synthesised to develop novel design recommendations for the information required inside Level 2 vehicles, and how this should be adapted over time depending on the driver’s familiarity with the system and driving events. This doctorate has contributed novel design recommendations for Level 2 vehicles through an innovative methodological approach across four studies. These design recommendations can now be taken forward to design and test new HMIs that can create a better, safer experience for future automated vehicles.
Due to the complexity of real driving and operating conditions of Battery Electric Vehicles, it is challenging to accurately estimate the remaining driving range of the vehicle. Relying only on traditional energy consumption prediction based on the historical data shows obvious low-fidelity and hysteresis, especially when the traffic route is unknown. The accuracy of future travel energy consumption prediction fails to be guaranteed once the switching of operating conditions is involved. For this reason, a map named “Driving Route Planning” Application Programming Interface server is built on the cloud, receiving online traffic route information, and the Hidden Markov Model is applied for prediction optimization of future operating conditions. The remaining driving range of Battery Electric Vehicles is finally estimated according to the future energy consumption and the remaining dischargeable energy. The results show that the prediction of operating condition combined with traffic route information and Hidden Markov Model reflects the switching of future operating conditions more accurately and quickly. The relative error of the remaining driving range estimation proposed keeps within 5% under the real operating verification.
There is an increasing need for charging station recommendation to minimize the overall charging time for electric vehicles and balance load for the charging stations. To grant this need, we model the recommendation problem as a Markov Decision Process (MDP) problem. However, the traditional MDP model has the issue of 'curse of dimensionality'. To address this issue, we propose an extension of MDP: multiple-phase MDP, in which the state transition of MDP is decomposing into several phases, so as to reduce the state space and state transition complexities. This is done by introducing two states other than the normal state defined in MDP: post decision state and intermediate decision state. Then, we propose an online learning based algorithm to solve the formulated multiple-phase MDP model. Thanks to the reduced complexities of the state space and state transition, the proposed online algorithm can converge fast. By comparing to other recommendation mechanisms, such as game theory based recommendation and Q-learning based recommendation, our simulation evaluation demonstrates that our proposition can bring good performance.
Conference Paper
Full-text available
This paper sets out to discuss how barriers to the uptake of electric vehicles (EVs) can be overcome using intelligent transport systems and will draw up a roadmap and action plan for the implementation of those services. Both, the roadmap and action plan have been designed in collaboration with the ERITCO Electromobility Task Force and the smartCEM project partners. The paper will discuss the need for future regulations and standards and take a strategic view on which stakeholders need to take a lead for the development and production of those services.
Full-text available
This study develops a more accurate range prediction for electric vehicles (EVs) resulting in a routing system that could extend the driving range of EVs through calculating the minimum energy route to a destination, based on topography and traffic conditions of the road network. Energy expenditure of EVs under different conditions is derived using high-resolution real-world data from the SwitchEV trial. The SwitchEV trial has recorded the second-by-second driving events of 44 all-electric vehicles covering a distance of over 400 000 miles across the North East of England, between March 2010 and May 2013. Linear models are used to determine the energy expenditure equations and Dijkstra's graph search algorithm is used to find the route minimising energy consumption. The results from this study are being used to better inform the decisions of the smart navigation and eco-driving assist systems in EVs, thus improving the intelligent transport systems provisions for EV drivers. The outputs of the research are twofold: providing more accurate estimations of available range and supporting drivers' optimisation of energy consumption and as a result extending their driving range. Both outputs could help mitigate range anxiety and make EVs a more attractive proposition to potential customers.
Conference Paper
This paper reports on the results from the EU-project ELVIRE regarding the phenomenon of range anxiety, the fear of not reaching the destination in an electrical vehicle (EV). Presented is the initial analysis of a questionnaire study with 42 Swedish EV drivers targeting their experience of range from a behavioural perspective (e.g., driving habits), as well as from a psychological perspective (e.g., experience of unease while driving). The following questions are addressed: (a) How is range anxiety portrayed by EV drivers? and (b) What is the role of experience of EVs for range anxiety? Regarding (a), findings show that the drivers’ posses a behaviour indicating that they suffer from range anxiety, although, they explicitly state that they have a low level of range anxiety; highlighting the unconscious nature of range anxiety. Regarding (b), statistical analysis show that there is a significant difference in level of experienced range anxiety and understandings of vehicle information but no significant difference in driving style between drivers who has more than 6-months experience compared to those with less than 6-month experience; indicating that driving style are manifested early on and stays stable over time.
Conference Paper
One challenge facing the adoption of electric vehicles (EVs) is the reduction of the impact of running out of fuel. An EV, with its limited charging point infrastructure and long charge times, is not seen as being as reliable as a conventional car in this respect. To make EVs more acceptable, forward-looking predictive methods of calculating range need to be developed which also take into account opportunities to conserve or harvest energy, as well as environmental factors such as terrain and weather conditions. Using a well-established EV range simulator, this paper describes scenarios showing the limitations of relying on such an approach and the potential detrimental results to both the driver and vehicle's ability to start and complete a journey. It provides an overview of the research being undertaken by the authors to address these problems, including a description of `Electrikitty', a road legal pure EV that will be used to gather data to verify the development of novel range estimation algorithms.
Conference Paper
The performance of a plug-in hybrid electric vehicle (PHEV) or an electric vehicle (EV) is closely related to the performance of its high-voltage battery pack. This is why, among the various means of maintaining a battery's peak performance, the importance of thermal management is emphasized. Since the early stages of the development of Li-ion batteries, much emphasis has been placed on battery cooling in order to secure stability, while less attention has been given to battery heating. However, in a low-temperature environment, using a battery without heating it severely undermines the battery's performance and life expectancy. A PHEV or an EV operates with energy that is stored by charging the battery, which can then be heated by an external source in a low-temperature environment prior to driving. In this study, a battery HILS and an environment simulation system are used to verify that pre-heating a battery in a low-temperature environment, using an external source, such as a charging stand, improves the battery's performance as compared to circumstances in which this is neglected.
This paper presents a Li-ion cell model parameterization technique for hybrid electric and electric vehicle control applications. The proposed method is based on an equivalent electrical circuit (EEC) model of the Li-ion cell and combines the advantages of the two main strategies employed for cell model parameterization, namely, the offline and online procedures. Offline methods are based on the identification of relevant EEC parameter values using a limited set of test data specific to the target cell chemistry. Conversely, online techniques employ adaptive algorithms that update the cell model as it is being used. The novel method presented in this paper employs recurrent offline updates of the EEC parameterization set, and thus, it integrates the advantages of the offline approach, such as flexibility, reduced complexity, and improved run-time performance, with the main benefit of the online counterpart, which is the capacity to adapt the model parameterization to uncharacterized operating conditions. Based on an extensive set of experimental and simulation results obtained from tests specified in the IEC 62660-1 standard, it is shown that the proposed approach offers a significant accuracy improvement over simple offline methods, as well as enhanced runtime speed in comparison with commonly employed online strategies.
Limited driving range is an obstacle to adoption of electric vehicles (EVs). We examine from a self-regulation perspective the psychological dynamics underlying individual reference values for three different types of range constructs. In a 6-month field trial 40 EVs were leased to a sample of early adopter customers. In general, users were satisfied with range and stressful range situations rarely occurred. Results further suggested that users were comfortable with utilizing approximately 75–80% of their available range resources. Several personality traits (e.g., control beliefs, low impulsivity) and system competence variables (e.g., daily practice, subjective competence) were positively related to range level values and thus range utilization. Comfortable range was positively related to range satisfaction. We recommend that psychology-based strategies should be applied to enhance range optimization.
The ultra-low carbon vehicle demonstrator programme -Final Report: Assessing the viability of EVs in daily life
  • S Carroll
  • C Walsh
  • M Burgess
  • M Harris
  • S Mansbridge
  • N King
  • L Bunce
S. Carroll, C.Walsh, M. Burgess, M. Harris, S. Mansbridge, N. King, L. Bunce, "The ultra-low carbon vehicle demonstrator programme -Final Report: Assessing the viability of EVs in daily life," Tech. Rep. Sept. 2013.
Human machine interfaces in low carbon vehicles: Market trends and user issues
  • T Wellings
  • J Binnersley
  • D Robertson
  • T Khan
T. Wellings, J. Binnersley, D. Robertson, T. Khan, "Human machine interfaces in low carbon vehicles: Market trends and user issues", Low Carbon Vehicle Technology Project, Technical Report: HMI 2.0, Dec. 2010
Electric vehicles: Improving consumer information to encourage adoption
  • N Wallis
  • B Lane
N. Wallis, B. Lane, "Electric vehicles: Improving consumer information to encourage adoption", European Council for Energy Efficient Economy, Tech. Rep. 4-514-13, 2013.