The Impact of Road Types on the Energy Consumption of Electric Vehicles

,


Introduction
Today, scarcity of nonrenewable energy resources and increasing ecological awareness have intensified efforts to transition to more environmentally friendly technologies.In transportation, the electrification of vehicles is a promising technology for greener, low-emission transportation systems.
e global electric vehicle (EV) stock has significantly increased over the last decade.In 2020, more than 10 million EVs were registered worldwide, 20% of which were in the United States (U.S.), making it one of the biggest EV markets in the world [1].Among EVs, battery electric vehicles (BEVs) have become increasingly popular and currently represent more than 75% of U.S. EV sales, and the growing demand for BEVs is expected to continue well into the future [1].
Nevertheless, range limitations and an insufficient charging infrastructure feed range anxiety (i.e., drivers' fear of running out of charge before reaching their destination) and impede BEV adoption [2].In addition to equipping EV models with larger batteries, one way of mitigating these limitations is adjusting driving behavior and optimizing BEV-specific routing.In other words, the range of BEVs can be extended by minimizing energy consumption through eco-driving and route choices.Understanding how operating characteristics on different road types impact energy consumption can be a practical and beneficial consideration in optimizing the energy-based routing of EVs.Knowing how much energy the BEVs consume in different driving situations is a prerequisite for range optimization.Findings of economical driving behavior can be directly implemented by drivers and further be used in vehicle navigation to identify the most economical route ("eco-routing") [3].Table 1 summarizes key findings of studies relevant to this paper.
To date, most of the literature investigating the impact of traffic planning and road types on fuel consumption and driving behavior has focused on conventional internal combustion engine vehicles (ICEVs) [4,5].Ericsson et al. [3], and Brundell-Freij and Ericsson [4] discovered that the road and traffic environment, the driver demographics (e.g., age and gender), and the car mass affect driving behavior.However, the studies lack a quantification of the road and traffic impacts on fuel consumption.Hallmark et al. [6] identified different on-road geometric and operational variables at signalized intersections and explored their effect on the time an ICEV spends in a specific operating mode.
e findings aimed to improve emission estimates for passenger vehicles but lack the quantification of fuel consumption.In Faria et al. [7], the fuel-saving potentials of ICEVs based on driving behavior, road types, and road grade were assessed.e research was conducted under natural conditions and monitored 47 drivers in Lisbon, Portugal over six months.
e highest fuel-saving potential was observed on minor arterial roads and collector roads with a speed limit of 50 km/h (31 mph).However, the study did not include BEV consumption models.Ericsson [8] investigated the variability in urban driving patterns and concluded that the street type has the most significant influence on the driving pattern for ICEVs.Furthermore, the study discovered that the average speed and deviation of speed are significantly different between different road types.However, specific consumption factors were not considered nor quantified.Jensen [9] evaluated different driving patterns and the resulting emissions of ICEVs on different road types.e author concluded that speed is the most critical factor, while the speed deviation in traffic did not explain fuel consumption.Slightly higher emissions were discovered on express roads compared to motorways.
Meanwhile, findings regarding ICEV consumption do not necessarily apply to BEVs.More specifically, BEVs' speed-related energy consumption is fundamentally different compared to ICEVs.According to Galvin's model, the minimum energy consumption of BEVs is at approximately 37 mph before increasing exponentially [10].ese findings were later supported by Fiori et al. [11] and Liu et al. [12], showing that BEVs reach their highest efficiency at lower speeds than ICEVs.Speed and its variation differ significantly between road types.erefore, road types are expected to be a significant factor in the analysis of assessing the depletion rate of BEVs [13].In conventional vehicles, frequent braking and accelerating significantly increase fuel consumption compared to trips with a steady speed, such as driving on highways or the interstate [7]. is effect could be different in electric vehicles because of the ability to recharge while decelerating [14].Additionally, BEV efficiency is more sensitive to ambient temperatures than ICEV fuel consumption.At low temperatures, BEVs are significantly less efficient than in warm temperatures due to higher battery storage losses [15,16] and internal resistance [17].erefore, different factors possibly impacting the energy consumption of EVs should be understood carefully.
Various studies have started this process using different approaches (see Table 1).For example, in Faria et al. [18], the energy consumption of three BEV models (i.e., Nissan Leaf, Smart ED, and Peugeot iOn) was investigated on a defined test track.In this study, the influence of the air conditioning system (i.e., off, on-cooling, on-heating), the driving style (i.e., aggressive, moderate), and the degree of urbanicity (i.e., urban, rural) were examined as influencing factors of energy consumption.An aggressive driving style increased the power consumption of the Nissan Leaf by 48% (25.01 kWh/ 100 mi to 16.85 kWh/100 mi).With a regular driving style and the air conditioning switched off, the Nissan Leaf consumed 21.1 kWh/100 mi on the combined route.However, cabin heating increased the energy consumption to 29.5 kWh/100 mi.Faria et al. [18] demonstrated the negative effect of aggressive driving behavior and quantified its influence on energy consumption in a naturalistic driving setting.Franke et al. [19] supported these findings by investigating the effects of the application of eco-driving strategies on BEV efficiency.e authors stated that driving economically has significantly reduced BEV energy consumption [19].Kurien et al. [20] modeled BEV energy consumption using acceleration and slope as input parameters in MATLAB Simulink.e results show that BEV efficiency decreases with speed and an increasing slope.In other words, high slope angles and traveling at high speeds were found to substantially limit the achievable ranges of EVs.
However, few studies have investigated the impact of road types specifically on BEV energy consumption.Applied road classifications often differ between these studies due to differences in the underlying definitions by municipalities, states, or countries.Yao et al. [14] collected data from road networks in Beijing to estimate EV energy consumption and emission factors using vehicle-specific power based on speed and acceleration.e authors discovered that energy consumption on principal arterial roadways was significantly higher than on other road types.In Wu et al. [21], data were collected through a BEV conversion driven by a single university faculty member over a total period of five months in California.e authors posited that in-city driving requires more energy than freeway driving.However, Wu et al. [21] differentiated all trips by these urban categories and did not account for driver behavior.Another study by Bingham et al. [22] investigated data collected from different drivers in a test vehicle on urban and country roads in the United Kingdom.
e authors discovered that aggressive driving could increase the average energy consumption by 30% or more.ey further stated that traffic management and reducing quick acceleration and deceleration offer additional energy savings and range extension potential.Bingham et al. [22] did not specifically address a correlation between road types and energy consumption in their work.One study that directly explored the impact of road types was conducted by Fetene et al. [23].e authors collected data from 741 BEV drivers over two years in Denmark to understand BEV efficiency behavior.No evidence was found that highway and non-highway driving were related to significantly different consumption rates.While road types were categorized in general terms, the study did not include a detailed definition and differentiation.
Consequently, current literature is associated with limitations when analyzing the impact of road types on EBV efficiency.While previous studies on the consumption characteristics of EVs suggested significant differences in energy consumption behavior over road types, they lack a clear definition of road types.
e energy efficiency parameters for distinct sections in road networks are necessary for modeling and planning an energy-optimal route.EV drivers already tend to choose certain road types associated with lower consumption levels [24], yet this behavior is based on assumptions and is not proven scientifically in a naturalistic setting.Furthermore, most of the literature findings were based on simulations.Manufacturer statements regarding energy consumption and range are measured under laboratory conditions for specified driving cycles and, thus, scientifically deviate from real-world consumption [25,26].Hence, there is currently a gap in the literature regarding identifying and understanding the energy consumption of EVs and BEVs in a naturalistic setting with human drivers.is study aims to close this gap by contributing towards a better understanding of BEV energy consumption on carefully differentiated road types.rough various regression models, this study investigates the impact of road types on the energy consumption of BEVs while accounting for other explanatory variables (i.e., ambient temperature, the initial state of charge (SOC), and driver).
e remainder of the paper will explore answers to the following research questions: Does the road type impact BEV energy consumption?If so, how do road types affect the efficiency of BEVs?With answers to these questions, vehicle operators can estimate the total cost of ownership associated with energy consumption based on vehicle usage. is is an essential element in the purchase decision of EVs and all future infrastructure decisions [2].

Materials and Methods
For this study, driving data were collected in a field study with a sample of human drivers.is section outlines the experimental design before describing data processing procedures.e section ends with a structured approach to the statistical analysis.

Experimental Design.
For a substantial enough sample size to detect the consumption behavior over different road types, a total of 30 study participants were recruited, following previous research that investigated driving behavior and energy consumption rates [27,28].
e driver characteristics were intentionally held as homogeneous as possible to mitigate the effect of individual differences in driving style on the average energy consumption of the vehicle.All participants were students at the university without prior experience in driving a BEV to exclude differences in driving behavior through different levels of BEV adaptation [19].
e participants were 40% female and 60% male, between the ages of 19 and 30, with a mean age of 23.73 years and a standard deviation of 2.56 years.All participants drove the test vehicle on two test routes that collectively covered all road types.Differentiation of the road types was undertaken accordingly to the U.S. Highway Functional Classification System of the Division of Statewide Planning, which defines the roads' purpose along with specific road characteristics, such as the material of a street, road width, traffic volume, location, and its function [29].e Division of Statewide Planning distinguishes between two primary functions of a roadway: (a) access to the property and (b) travel mobility [29]."Mobility refers to the actual ability of the road to move traffic, while accessibility refers to the ease of entering or exiting a roadway to or from adjacent priorities" [29].Accordingly, the U.S. road network is divided into three major road classes: arterials, collectors, and local roads.e division defines these as follows: "arterials have high mobility but low land access and are typically used for longer trips whereas local roads have low mobility (travel at slower speeds) but provide the highest level of land access.ese are used for shorter trips around town.Collectors are in the middle and often act as the transitional roads from arterials to local roads" [29].Furthermore, each road class is subdivided into multiple road types.Table 2 lists the three road classes and the respective road types.e road type "interstate" serves high-speed travel and long-distance freight transport by motor vehicle (high mobility) and connects the main urban areas of the U.S. By definition, "other freeways and expressways" include multiple lanes in both directions that are typically separated by a physical barrier (e.g., traffic barrier, a strip of grass, or a boulder) [29].e main function of a "principal arterial," also known as a traffic artery, is to direct traffic between "collectors" and "other freeways and expressways" and ensure a high level of service between major urban and metropolitan areas [29].erefore, access to arterial roads is often limited and only possible through designated roads."Minor arterial" roads serve trips of moderate length and at lower speeds.In rural areas, their purpose is to link cities and larger towns while also providing land access to the surrounding areas.Collector roads serve shorter distances at lower speeds than arterial roads.Typically, "minor collectors" are shorter, with a higher density of local driveways and lower speed limits."Major collectors" might have multiple travel lanes and serve a higher annual traffic volume."Local roads" offer direct access to abutting land and are often designed to hinder traffic.ese roads make up the most significant proportion of all roads within the U.S. but are not intended to serve long-distance travel.erefore, "local roads" are usually traveled at the beginning and end of a trip.
Two test routes were differentiated to obtain driving data from all road types: (i) Route A and (ii) Route B. e road types covered by Route A were "other freeways and expressways," "other principal arterial," "minor arterial," "major collector," "minor collector," and "local roads."To include the road type "interstate" and avoid the risk of entering a critical SOC through excessive distances, the research was extended by Route B, which included the road type "interstate."Due to unforeseen construction on a road section on Route A, the road type "minor collector" had to be excluded from the analysis.Table 3 shows the distances covered on each test route and road type.
Based on EV driver travel patterns [30,31] and research on air pollution [32,33], peak hours for travel were assumed between 7 : 00 AM and 9 : 00 AM and after 4 : 00 PM. ese times were avoided to mitigate the impact of commuter traffic on the data.Accordingly, weekends and holidays were excluded due to possible differences in traffic density.Street surface conditions, bad visibility, and darkness significantly influence the vehicle's energy consumption [34,35].Darkness was avoided, and trials with wet roads or low visibility were rescheduled.Data were collected in April and May of 2019. is short trial period was chosen to minimize the outside temperature range and avoid seasonal temperature effects [23].Air conditioning settings were held constant.e windows remained closed to keep the aerodynamic drag the same throughout each test drive.e test vehicle, a 2017 Volkswagen eGolf, is equipped with a 35.8 kWh battery pack and offers multiple driving modes (i.e., Normal, Eco, and Eco+) and recuperation intensities (i.e., B, D, D1, D2, and D3) [36,37], which in turn have a  [38,39].In this research, the driving mode "normal" was used since it uses the full power of the electric engine and, therefore, makes the best use of the vehicle's performance.Additionally, the recuperation mode "D" was used, which recuperated energy only while applying the brakes.Since the participants in this study were first-time EV drivers, they were given a brief introduction to the car before the test drive.ey were provided with the opportunity to adjust the seat to their comfort for safety reasons.e same playlist was played during the test drives to exclude differences in behavior due to the music being played [40].

Methods of Data Collection and Data Processing.
e test vehicle was equipped with a controller area network (CAN) bus gateway with a data logger to collect energy consumption data.e CAN bus gateway is a device offered by the company FleetCarma which allows for the extraction of the corresponding data from the vehicle's onboard diagnostics-II system and links it to the corresponding GPS location and time [41].e resulting data included information on the vehicle's geographic location, battery current, battery voltage, and speed at a rate of 1 Hz.e raw data were processed in Python through project-specific code using the Pandas [42] and Matplotlib [43] packages.Before data processing, the raw data were cleaned to ensure comparability and uniformity of each trial.e start and endpoint of each test route and the coordinates of each new road type were defined.Additionally, data before the starting point and ending point were deleted.A perfect match of the GPS location and the location of the drives was rarely achieved due to minor deviations of every drive and the GPS tracking method.Due to the previously mentioned construction on Route A, this 1.9-mile stretch was removed from the data.e total energy consumption for each road type of every trial could be calculated based on the battery current and voltage.e formula of uniform motion relates information about distance, time, velocity, and initial distance to each other and led to the total distance driven for every trial.

Methods of Data Analysis.
e dependent variable in this study was the average energy consumption per mile. is continuous variable could be expressed as the difference in the state of charge per distance (ΔSOC/distance) and was measured in kilowatt-hours per mile (kWh/mile).e explanatory variables considered consisted of continuous variables (i.e., "initial SOC" and "ambient temperature") and categorical variables ("driver" and "road type").Linear regression models allow for analyzing the effect of categorical and continuous variables on a continuous dependent variable and have been used for analyzing EV efficiency in the past [37].
erefore, the linear regression analysis was chosen to examine the impact of the considered explanatory variables on the average energy consumption per mile.e necessary variables for the analysis and the computation method are summarized in Table 4.
To determine whether different road types impact the energy consumption of EVs, all six road types (see Table 3) were evaluated on their distinct influence on BEV energy consumption.After the datasets for all drives on both test routes were separated into the different road types, a sample size of n � 390 data points was achieved.e continuous variables "temperature" and "initial SOC" were considered in the model.To analyze the impact of categorical variables "road type" and "driver," dummy variables were used in the regression models.Since differences in driver behavior significantly influence the energy consumption rate of vehicles, the characteristics of the participants were chosen to be as homogeneous as possible to minimize the effects of individual driver behavior.
e factor "driver" was still included in the models as an independent categorical control variable.Drivers were numbered randomly from 1 through 30.e driver with the average consumption closest to the sample average on both test routes was chosen as a reference for the driver dummy variable.As the most consistent road type in lane and shoulder width, median separation, and entrance/exit ramps, the road type "interstate" served as reference [44].A stepwise selection was then applied to find a potentially better fitting model.In addition to considering all variables separately, additional regression models accounting for interaction effects between particular variables and "road type" were analyzed and compared.ese models accounted for an interaction effect between "driver" and "road type," "temperature," and "road type," and "speed" and "road type," respectively.e impact of speed has been shown in various studies [10][11][12]20] and was, therefore, included in the analysis.
All regression models were tested for the observance of the different prerequisites of linear regressions (i.e., no outliers, independence of residuals, homoscedasticity, and normality of residuals).An analysis of variance (ANOVA) table and a detailed explanation of the results are only given for regression models that fulfilled all the mentioned prerequisites.e detection of potential outliers was carried out through the Grubbs' test [45].Normality testing was performed through histograms and scatterplots of the quantiles of the data.Under consideration of the sample size, the Jarque-Bera test statistic was calculated to support normality testing mathematically.To analyze real-world data, a similar study by Bartels et al. [36] chose a significance level of α � 0.1.To reduce the risk of a Type I error, we decided to use a significance level of α � 0.05 for all statistical tests.e Journal of Advanced Transportation coefficient of determination (R 2 -adj.) was calculated and interpreted for all models according to the guidelines of Cohen [46] to assess all regression models.R 2 values for models with endogenous latent variables can be divided into a weak explanation of variance (0.02 ≤ R 2 < 0.13), moderate explanation of variance (0.13 ≤ R 2 < 0.26), and strong explanation of variance (R 2 ≥ 0.26).

Results
Before carrying out the regression models, the prerequisites listed in the previous section were checked.Although one data point for the road type "local road" appeared to be significantly higher with an average consumption of 0.39 kWh/mile, the Grubbs' test for outliers did not support the appearance of an outlier (p � 0.671).erefore, no data points were removed from the data set.According to both the histogram of the data (Figure 1(a)) and the scatter plot of sample quantiles versus theoretical quantiles (Figure 1(b)), the data appeared to be approximately normally distributed.
In addition, the Jarque-Bera test statistic supported evidence for normally distributed data (p � 0.068).Table 5 summarizes the descriptive statistics for the average energy consumption per mile, the speed in miles per hour (mph), and the variation in speed in mph for all considered road types.e results match presumed speeds on some road types considering their definition.e highest average and median speed were found on "interstate," followed by "other freeways and expressways" and "other principal arterial."In contrast, "local road" and "major collector" showed average and median speeds than "minor arterial," which was related to the second-highest variation in speed.A comparatively high variation in speed was discovered on "interstate" and "other freeways and expressways," which are typically related to steady speeds.e results further indicated that mean, median, and maximum average energy consumption were comparatively higher on "interstate" and "local road."Meanwhile, the road type "major collector" had the lowest minimum, median, and mean energy consumption with 0.0616 kWh/mile, 0.1486 kWh/mile, and 0.1608 kWh/mile, respectively.e highest minimum average consumption per mile was on the "interstate" with 0.1987 kWh/mile.Data from the road types "major collector" and "local road" showed higher variation than other road types with a standard deviation of 0.0685 kWh/mile and 0.0683 kWh/mile, respectively.is is also shown in Figure 2. e plot further shows that the road types "other freeways and expressways" and "other principal arterial" have more minor variations than some other road types.
e results for the first regression model that considered all factors separately without any interaction effects are summarized in Table 6.
While the factors "driver" (p � 0.774) and "initial SOC" (p � 0.065) did not show any significance under the chosen significance level, "temperature" (p � 0.040) and "road type" (p ≤ 0.001) impacted the energy consumption of the vehicle significantly.Additionally, all different road types showed statistical significance.is model had a coefficient of determination of 35.95%.According to the Jarque-Bera test for normality, there was evidence that the residuals were following a normal distribution (p � 0.088).e plot of the residuals versus fitted values is displayed in Figure 3.It provides further evidence of a linear relationship between the dependent and the independent variables.Furthermore, the residuals are randomly distributed along the zero line, which supports the assumption of homoscedasticity.
An extended ANOVA table is summarized in Table 7.All road types are significant for the model when considered separately.Negative coefficients of all road-type dummy variables provide evidence that driving on the road type "interstate" is related to the highest consumption with a coefficient of 0.374 kWh/mile compared to driving on any other considered road type.For example, driving on "local roads" would reduce the average consumption per mile by 0.0284 kWh/mile, and driving on the road type "major collector" by 0.1093 kWh/mile.e road types "minor arterial" and "other principal arterial" have a 0.0647 kWh/ mile and 0.0654 kWh/mile lower consumption than "interstate."e factors "initial SOC" and "driver" were removed from the initial model through a stepwise regression process.e significant factors "temperature" and "road type" remained explanatory variables.e results provided evidence of the significance of the factor "road type" for the dependent variable (p ≤ 0.001).Although remaining in the model, "temperature" did not show the significance for BEV energy consumption (p � 0.136).is model had an R 2 -adj. of 37.06%.However, the residuals did not follow a normal distribution (p � 0.024).Due to the violation of this

Road type Categorical
Road types were differentiated according to the definitions of the division of Statewide Planning [29].Temperature °F Average ambient temperature information at 1 Hz.6 Journal of Advanced Transportation criterium of linear regression, the results for this model will not be further explained.ree more regressions were modeled, which accounted for possible interactions between particular variables and "road type."All of these models included the continuous variables "temperature" and "initial SOC."A model that included the interaction between "driver" and "road type" adds one dummy variable for each driver and road type combination, leading to a total of 180 [30 * 6] combinations.However, the coefficient of determination of this model was R 2 -adj.� 1.169%.e residuals did not follow a normal distribution based on the Jarque-Bera test (p � 0.010).e model will, therefore, not be considered further.
Table 8 summarizes the ANOVA results for the regression model that accounted for interaction between "temperature" and "road type."e results showed that the interaction effect "temperature" and "road type" is significant for the vehicle's mean energy consumption (p ≤ 0.001).e factors "driver" (p � 0.815) and "initial SOC" (p � 0.085) did not show the significance for the dependent variable.
is regression model explained 35.33% of the variance.
e Jarque-Bera test statistic indicated that the residuals followed a normal distribution (p � 0.1652).
Figure 4 plots the model's residuals versus fitted values that account for an interaction effect between "temperature" and "road type."e residuals occurred randomly along the zero line, which supports the assumption of a linear relationship.ere is also no indication for heteroscedasticity of the residuals.
An extended ANOVA table for this model can be found in Table 9, displaying p-values for all "road type" dummies and their coefficients.e respective interaction effects of "temperature" with the road types "major collector" (p � 0.021), "minor arterial" (p � 0.040), and "other principal arterial" (p � 0.029) were significant.e highest effect on the dependent variable was caused by an interaction    between "temperature" and "major collector."Accordingly, an increase in "temperature" by 1 °F would decrease the energy consumption by 0.0024 kWh/mile.In contrast, "temperature" and "road type" have the lowest effect when driving on an "interstate" with a decrease in consumption of 0.006 kWh/mile for an increase in "temperature" by 1 °F.e model led to equal coefficients of −0.0024 for "temperature" interactions with "minor arterial" and "other principal arterial."e coefficients of "temperature" and "local road," and "temperature" and "other freeways and expressways" were given as −0010 and −0012, respectively.is shows that the effect of temperature is lower on road types related to a higher general consumption compared to the first model.A final regression model was analyzed accounting for an interaction effect between "speed."Although leading to the highest coefficient of determination with an R 2 -adj. of 0.4976, the residuals of this model did not follow a normal distribution (p ≤ 0.001).
erefore, the condition of normally distributed residuals was not met, and the model could not be considered for the analysis.

Discussion
Different regression models were used to investigate the impact of road type on electric vehicle efficiency.All explanatory variables (i.e., "initial SOC," "driver," "temperature," and "road type") were initially considered separately.
is first model showed the highest explanation of variance with a coefficient of determination of 35.95%.Furthermore, there was evidence for homoscedasticity of the data.e model discovered that both the factor "temperature" and the factor "road type" are significant for the energy consumption of electric vehicles.Road types, in general, appear to impact BEV efficiency significantly.Accordingly, a fully charged battery operating at a battery service life optimal ambient temperature of 68 °F [16,17] driving on an "interstate" would lead to the highest energy consumption.Figure 5 visualizes the consumption behavior under the given circumstances for all drivers on different road types.
e black line displays the mean of the fitted values for all drivers.Driving on an "interstate" would lead to an average consumption of 0.2367 kWh/mile in the applied setting.e negative regression coefficients have shown that driving on any other considered road type would increase the vehicle's efficiency.For example, driving on a "local road" as the second-largest energy-consuming road type would lead to an average consumption of 0.2082 kWh/mile.Ceteris paribus, the lowest average consumption, was related to driving on a "major collector," which was associated with a consumption of 0.1273 kWh/mile on average among all drivers.
e remaining considered road types, "other freeways and expressways," "other principal arterial," and "Minor Arterial," show similarity in terms of means of the fitted values.Table 10 summarizes the relative change in mean energy consumption from one road type based on the fitted values of the first model.BEVs operate most efficiently on "major collectors" with a savings potential of 46% when choosing this road type over an "interstate."Avoiding the "interstate" in all cases leads to a reduction in consumption per mile of at least 12%.Avoiding "local roads" can save between 5% and 38% if choosing an alternate road type that is not an "interstate."e difference in efficiency between "minor arterial" and "other principal arterial" is only marginal and does not appear to be significant when choosing one over the other.
While this study differentiated between road types more carefully than previous studies, the results align with findings in [21], which found that in-city driving is related to higher consumption levels than freeway driving, assuming that in-city driving means a high ratio of "local roads."e relatively high average consumption level on "local roads" might be caused by the high variation in speed on these roads, which are often interrupted by intersections.In Bingham et al. [22], it was stated previously that less acceleration and deceleration could reduce the average consumption of BEVs.Since one of the road type classification criteria is the ability of a road to move traffic [29], the relatively high consumption on "local roads" is consistent with Bingham et al.'s findings.In other words, "local roads" are not designed to move high traffic volumes while ensuring uninterrupted flow.
is study could not confirm previous findings by Yao et al. [14], who found in their study that BEVs perform the least efficiently on principal arterial roadways.While their underlying definition of this type of road might differ, "other principal arterials" were not related to the highest consumption in this study.e results provided evidence that "interstates" in particular lead to high consumption, which contradicts the results of [23].e authors posited that a  Journal of Advanced Transportation significant difference between highway and nonhighway driving could not be found.Again, the underlying definitions used in [23] and this study might differ and could be caused by general differences in classification criteria between the U.S. and Denmark.Furthermore, the study showed that BEV discharge behavior could be fundamentally different from ICEV fuel consumption behavior.Notably, while ICEVs perform more efficiently on road types with steady speeds (e.g., highways and interstates) [7], BEVs appear to behave more efficiently on different road types (i.e., "principal arterials" and "minor arterials").e only road type where ICEVs and BEVs are relatively efficient is on "major collectors" [7].
While the results of this study have provided evidence that BEVs' discharge behavior depends on the used type of road, other factors impact BEV efficiency.More specifically, with an increase in ambient temperature, BEVs operate more efficiently.is behavior aligns with studies on battery performance in different temperatures [15][16][17].Batteries are, in general, less efficient in extreme temperatures [47].e factor, "initial SOC," was insignificant as previously also shown in [37].e control variable "driver" was not found to significantly impact the energy consumption of electric vehicles.is variable was used as a control variable.Its insignificance was likely due to the homogeneous driver demographics of the sample (all participants were students under the age of 30 and had no previous experience in BEV driving).is result aimed to mitigate the effect of variation in driving styles.It should be stated that previous studies demonstrated the impact of driver behavior (e.g., aggressive versus nonaggressive) on BEV efficiency [18,19,48].Furthermore, the impact of traffic congestion on BEV consumption behavior was minimized to the extent possible by controlling for the  However, it should be mentioned that the negative effect of traffic on BEV performance was previously discovered in Jonas et al. [49].A stepwise selection did not lead to a better model because the residuals for the model without the factors "driver" and "initial SOC" did not follow a normal distribution.is was also found in the model accounting for the interaction between "road type" and "driver."In contrast, the model including the interaction between "temperature" and "road type" led to residuals following a normal distribution.us, the model could explain almost the same percentage of variance as the first model.e model demonstrates that individual interactions between the road types "minor arterial," "major collector," and "other principal with the factor "temperature" are individually significant for BEV discharge behavior.More specifically, the effect of temperature varies with the road type.Accordingly, the lowest energy savings potential per 1 °F increase in ambient temperature can be observed on "interstates" (−0.006 kWh/mile), followed by "local roads" (−0.0010 kWh/mile), "other freeways and expressways," and "other principal arterials," and "minor arterials" (both −0.0017 kWh/mile).Finally, the highest savings potential was found for "major collectors" with a reduction in average energy consumption of 0.0024 kWh/mile per 1 °F increase in ambient temperature.
Both models demonstrated the importance of road type for the efficiency of BEVs and contributed to a deeper understanding of BEV consumption behavior.Future studies aiming to improve BEV-specific routing and establish consumption estimation models should consider road types as an essential variable.Drivers can consider these results in improving their eco-driving behavior and, hence, extending achievable ranges significantly by adjusting their route choices accordingly.Furthermore, the results of this study can be implemented into smart routing applications to increase the eco-driving potential.Adjustments to routing made by drivers and service providers can reduce the impeding effect of range anxiety and contribute to BEV adoption.

Conclusion
Understanding the consumption behavior of BEVs and driving factors is essential to reducing the barriers to widespread adoption.is study is the first to implement an analysis that captures the effect of carefully differentiated road types on the energy consumption of BEVs while using real-world driving data by a sample of BEVinexperienced drivers.Multiple models have been analyzed and compared.We discovered that the type of road is crucial for the BEV consumption of BEVs.Considering the type of road in driver's route choices and BEV-specific  Journal of Advanced Transportation smart routing applications offers significant energy savings potential of up to 46%.We discovered that the highest consumption levels are related to driving on "interstates" and "local roads."e performance of BEVs was found to be significantly less efficient on roads that allow for highspeed traveling (i.e., "interstates") and roads with deficient mobility that interrupt traffic regularly (i.e., "local roads").e results provide strong evidence of the importance of route choice on BEV efficiency.
However, there are limitations to the study.e road type "minor collector" could not be considered in this study due to unforeseen construction.Further research is needed to determine the effect of this specific road type on energy consumption.ere was no slope data available for the test drives, meaning the effect of elevation differences was not considered as a factor that may have affected the results.Future work should consider slope or elevation data since the interaction between the type of road and the slope is presumed but not currently quantified.Traffic data were not available for the test routes during data collection.Traffic volume and environment often correlate with the type of road.erefore, considering traffic and intersection measures in future studies would be a beneficial extension of this study.Additionally, test runs were undertaken by a small and homogeneous sample of drivers.Testing with a broader and more diverse sample of drivers would account for differences in driving behavior.
e findings of this study can be used by service providers, academics, and BEV users to improve energy-optimal route planning and range forecasts.In addition, the practical consideration of these findings can help BEV drivers to successfully apply and optimize eco-driving strategies that extend achievable ranges.e results thereby contribute to a better understanding of BEV efficiency factors and facilitate the transition to low-emission transportation systems.

Figure 1 :
Figure 1: (a) Histogram and (b) Q-Q plot of average energy consumption per mile for the entire sample.

Figure 2 :
Figure 2: Jitter plot of mean energy consumptions for all road types.

Figure 4 :
Figure 4: Residuals versus fitted values plot for the model with interaction effect of road type and temperature.

Figure 5 :
Figure 5: Fitted values for drivers and different road types.

Table 1 :
Studies relevant for this paper by author, year, location, type of vehicle analyzed, and relevant key findings of the publications.

Table 2 :
Road classes and types.

Table 3 :
Distance covered on each road type.

Table 4 :
Variables and explanation of computation.

Table 5 :
Descriptive statistics of consumption data per mile by road type.

Table 6 :
ANOVA of the road type analysis.

Table 9 :
Extended ANOVA table of the road type analysis with interaction effect of road type and temperature.

Table 10 :
Energy consumption comparison matrix.