Simulative Consumption Analysis of an All-Electric Vehicle Fleet in an Urban Environment

Abstract

The increasing shift towards battery electric vehicles (BEVs) in urban environments raises the question of how real-world traffic conditions affect their energy consumption. While BEVs are expected to reduce local emissions, their total energy demand, particularly in city traffic with with low average speeds, and therefore a higher impact of secondary consumption, remains insufficiently understood. To address this, a simulative framework to analyze the average energy consumption of an all-electric vehicle fleet in a mid-sized city, using Darmstadt, Germany, as a case study, is presented. A validated microscopic traffic simulation is built based on 2024 data and enriched with representative powertrain models for various vehicle classes, including passenger cars, trucks, and buses. The simulation allows the assessment of consumption under different traffic densities and speeds, revealing the substantial influence of secondary consumers and traffic flow on total energy demand. Furthermore, the study compares the $C{O}_2$ emissions of an all-BEV fleet with those of a fully combustion-based fleet. The findings aim to highlight the role of secondary consumers in urban traffic and to identify the potential for energy-saving.

1. Introduction

In 2023, the European Union (EU) agreed that from 2035 onwards, newly registered vehicles must have a net $C{O}_2$ emission of zero. Besides special regulations regarding vehicles driven with e-fuels, this essentially means that possible vehicle powertrain technology is, by EU regulation, limited to battery electric vehicles (BEVs) [1].

As well as emissions in the production and recycling phases, $C{O}_2$ emissions during the use phase are notable, caused mainly by the drivetrain of the vehicle [2]. For driving BEVs, those $C{O}_2$ emissions are not local emissions, but caused by electric energy production and are dependent on the corresponding share of renewable energies in the existing power grid. In Germany, this share was 53% in 2024 [3]. Even though this number has increased during recent years, the goal to have an energy grid, which is only based on renewable energy, is not even expected until 2035 [4].

Hence, the electric consumption while driving with pure BEV fleets will still generate $C{O}_2$ emissions. As a result, there is potential for $C{O}_2$ emission reduction by analyzing the consumption composition for BEV fleets.

Due to increasing urbanization, by 2050, about 70% of people in Germany will live in cities, which shows an increase of 10% compared to 2025 [5]. Together with the fact that around 80% of people in medium size and smaller cities own a car [6], these regions require great attention when talking about vehicle fleets. Especially in these urban areas, a significant share of the consumption while driving is based on secondary consumption (SC) for BEVs [7]. BEVs’ driving consumption is relatively low in low-speed urban traffic compared to combustion engine vehicles, which has already been investigated in [8,9], and is commonly known; however, the impact of the SC on total driving consumption is widely unknown.

In this paper, the average consumption for an all-BEV fleet is investigated based on a validated traffic simulation of Darmstadt, Germany, which represents a medium-sized city. The average consumption of a vehicle in this traffic scenario is examined based on its cause (e.g., drivetrain and SC) in order to obtain insights of possible energy-saving potential. Also, the impact of traffic density is investigated with two different traffic scenarios, which are modeled with different traffic demands. Following this, a statement is made on the extent to which EU legislation reduces $C{O}_2$ emissions while driving compared to combustion engine technology. In addition, areas with further energy-saving potential are to be identified.

2. Traffic Simulation

To investigate the consumption of an urban vehicle fleet, for this work, a traffic simulation of the city center of Darmstadt is generated, which is based on 2024 traffic data. To also investigate different traffic conditions, two different simulations are generated: a simulation with average traffic volume of Darmstadt and another with rush-hour traffic volume. The simulations include trucks, and passenger cars (PAC). Other traffic participants are not included because their impact on energy consumption and on traffic behavior is negligible. The simulations contain around 1 h of steady traffic behavior, with a start-up and a wind-down phase. The number of vehicles over simulation time for both simulations is displayed in Figure 1.

These traffic simulations are executed in the framework of “Simulation of Urban Mobility” (SUMO) [10]. SUMO is a microscopic, internal, multimodal and time-discrete traffic flow platform. These features make it possible to model each vehicle individually and analyze its driving behavior. SUMO is already common for use in research regarding traffic behavior, traffic effects of vehicle movements and for potential analysis regarding autonomous traffic. For example, in [11], SUMO is used to investigate traffic management strategies for pollution reduction. Therefore, a simulation of the city center of Braunschweig, Germany, is used. In [12], SUMO is used to investigate traffic flow optimization based on traffic light phases. In regard to autonomous traffic, Eichenlaub [13] uses SUMO to investigate an efficiency-optimized longitudinal control approach.

To get valid results from a traffic and vehicle simulation, a realistic driver and vehicle behavior is needed. To comply, SUMO uses different modeling approaches regarding driver behavior and traffic infrastructure, which can be parameterized. Furthermore, the traffic environment is crucial for realistic traffic simulation. Therefore, SUMO uses a traffic network approach, where intersections and streets are modeled as nodes and edges. In the following, the build up of the traffic network and infrastructure, the parameterization of the individual vehicle parameters, and the modeling of the traffic demand are presented.

2.1. Street Network

The simulations of the city center of Darmstadt contain the area which is shown in Figure 2. The simulation network, which is used for both considered simulations, was created using Webwizard, a tool provided within the SUMO framework. It supports the conversion of OpenStreetMap data [14] into SUMO-compatible formats via a web-based interface. With this, an accurate street network is built, which also includes traffic infrastructure, e.g., traffic lights and speed limits.

By building the street network with Webwizard, there is no elevation data included. To calculate accurate vehicle consumption, elevation data is significant, as it can be seen in Section 3.2. Therefore, the network is extended by adding this information from the “Hessian Administration for Land Management and Geoinformation” [15]. In Figure 2, the street network with this elevation information can be seen.

2.2. Car-Following-Model

The longitudinal motion of a driver–vehicle unit in SUMO is calculated with a car-following model. For this work, the Extended Intelligent Driver Model (EIDM) [16] is used. This model is based on the Intelligent Driver Model (IDM) [17] and is extended to build a more realistic driver behavior by adding, e.g., random driving errors. Basically, the speed v of a vehicle for a certain time is calculated with the acceleration $a_t$ by Equation (1).

$$v_{t+\Delta t}=v_t+a_{t+\Delta t}·\Delta t$$

(1)

The acceleration $a_t$ is therefore calculated by the execution in Equation (2). Here, a distinction is made between whether a dynamic safety distance $s_t^{*}$ is greater or less than the actual distance $s_t$ between a vehicle and the corresponding leader vehicle.

$$a_{t+\Delta t}=\left\{\begin{array}{cc}{a}_{\mathrm{free},t}·\left(1-{\left({\displaystyle \frac{s_t^{*}}{s_t}}\right)}^{{\displaystyle \frac{2a_{\max }}{|a_{\mathrm{free},t}|}}}\right),\hfill & \mathrm{for}{s}_t>s_t^{*}\hfill \\ \\ a_{\max }·\left(1-{\left({\displaystyle \frac{s_t^{*}}{s_t}}\right)}^2\right),\hfill & \mathrm{for}{s}_t\le s_t^{*}\hfill \end{array}\right.$$

(2)

Here, the maximum acceleration $a_{\max }$, which is defined in more detail in Section 2.3, for each vehicle individually is used. Also, the theoretical acceleration when traveling free $a_{\mathrm{free}}$ is used, as can be seen in Equation (3).

$$a_{\mathrm{free},t}=a_{\max }\left(1-{\left({\displaystyle \frac{v_t}{v_0}}\right)}^{\delta }\right).$$

(3)

In SUMO, the speed exponent is defined in SUMO with $\delta =2$ and the allowed speed is stated as $v_0$. The dynamic safety distance $s_t^{*}$ is calculated with Equation (4).

$$s_t^{*}=s_0+max\left(0,v_{\mathrm{lead},t}·T-{\displaystyle \frac{v_t·\Delta v_t}{2\sqrt{a_{\max }·\mathrm{b}}}}\right)$$

(4)

Here $s_0$ is the minimum allowed distance between two vehicles, T is the desired time gap, which is also explained in more detail in Section 2.3, as well as $\mathrm{b}$, which is the comfortable deceleration.

This calculation is additionally affected by uncertainty and perception errors, which is implemented by a Wiener process. The calculation of the Wiener process in SUMO is displayed in Equation (5).

$$w\left(t\right)=e^{-{\displaystyle \frac{\Delta t}{\tilde{\tau }}}}·w(t-1)+{\sigma }_{error}·\sqrt{{\displaystyle \frac{2\Delta t}{\tilde{\tau }}}}·\eta \left(t\right)$$

(5)

Here, $\tilde{\tau }$ is the correlation time of the processes, which is the inertia of the Wiener process. $\Delta t$ is the time step size, which is stated as $0.2$ s in this work. Additionally, $\eta \left(t\right)$ is a normally distributed random variable. The value of ${\sigma }_{error}$ is the variance of the distribution, which is further investigated in Section 2.3.4.

Furthermore, other minor adaptions, compared to the IDM, are executed regarding the driving behavior in certain traffic situations. These adjustments can be found in [16].

2.3. Vehicle Parameterization

In SUMO, the longitudinal motion of a vehicle is modeled as a driver–vehicle unit with the EIDM. This means that the customizable parameters of the EIDM need to be set in order to satisfy vehicle and driver characteristics and behaviors.

The different parameters can be influenced by the technical vehicle limits, but can also model inaccuracies and perception errors, which are characterizing driver behavior. To create a variety of different driver–vehicle units, the main parameters for the EIDM are chosen for each of the vehicles within the simulations based on a certain distribution. The distribution for each parameter is chosen based on relevant literature and is presented in the following part. Validation is done with real driving data.

2.3.1. Acceleration and Deceleration

There are various instances in the literature investigating the acceleration and deceleration behaviors of vehicles in urban traffic environments. For example, Burhoff and Lange [18] have investigated accelerations at pedestrian crosswalks and define them between $1.5$ m/s² and $2.2$ m/s² for vehicles, driving straight ahead. For turning vehicles, values between $1.0$ m/s² and $2.0$ m/s² are discovered.

In [19], the investigation does not further distinguish between traffic situations. Therefore, a general acceleration between $1.0$ m/s² and $2.5$ m/s² is recorded in urban traffic.

Omar et al. [20] investigate deceleration until stopping for vehicles driving between 30 km/h and 70 km/h. Here, values between $-0.6$ m/s² and $-1.0$ m/s² are recorded. It is also recorded that at higher speeds, braking time and distance increase but deceleration decreases.

For the EIDM car following model, the acceleration and deceleration are influenced by parameters, which are defined in SUMO as accel and decel. The parameter accel represents the acceleration limit and is therefore a fixed limit. The parameter decel is defined as the comfortable deceleration, which is a soft limit and can be exceeded depending on the traffic situation to maintain a safe drive.

In order to model the values defined in the literature, the accel and decel parameters are selected from a logarithmic normal distribution. The mean values and standard deviations are listed in Table 1.

2.3.2. Time Gap

Another parameter influencing the longitudinal motion of a vehicle with the EIDM is the time gap. This value defines the desired time distance between two vehicles and is the time which passes between two following vehicles after a certain point.The absolute length is therefore dependent on the speed of the vehicles. In [21], an approach is made to use real traffic on vehicles’ time gaps and transfer this into a simulation. The displayed results show a tendency for lower time gaps of around 2 s with higher speeds of around $12.5$ m/s compared to lower speeds of below 5 m/s, where time gaps around 4 s can be recorded.

According to this, the distribution for the value $tau$, which represents the time gap in SUMO, is selected from a logarithmic normal distribution. The values defining the distribution are shown in Table 1.

2.3.3. Reaction Time

The reaction time comprises the time from the occurrence of a stimulus to the occurrence of the resulting reaction. It also includes the motor action time of the vehicle and the reaction time of the driver. It depends on the driver’s characteristics such as age or physical or emotional state, and environmental conditions such as road or weather conditions [22].

A test to determine reaction time in urban traffic is described in [23]. In two consecutive vehicles, the driver of the rear vehicle was meant to press the brake pedal as soon as the brake light of the vehicle in front turned on. The frequencies of the determined reaction times corresponded to a logarithmic normal distribution with values between 0.3 s and 1.6 s. The values are therefore lower than in the distribution that was determined in [22]. Here, the reaction time was determined by measuring the time between the phase change of a traffic signal to green and the movement of the first vehicle. This resulted in logarithmically distributed frequencies of up to 6 s. The different results differ by the expectation of the driver, which is increased in the experiment of [23] and thus leads to a faster action.

In the EIDM, the reaction time is presented as the parameter treaction. This parameter defines the maximal time interval within a driver–vehicle unit can reassess a situation and perform an action. This value is therefore handled as a limit and not as a fixed value. The actual reaction time varies below this limit.

For selecting a range for this parameter, a logarithmic normal distribution with the parameters shown in Table 1 is used, based on the literature review. Here, $\mu$ and $\sigma$ denote the mean and standard deviation of a normal distribution. For the log-normal case, the starred parameters (${\mu }^{*},{\sigma }^{*}$) refer to the mean and standard deviation of the variable’s logarithm.

Table 1. Parameter distribution for cars.

Parameter	Vehicle Class	SUMO Default Value	Distribution	$\mathit{\mu }$ or ${\mathit{\mu }}^{\mathsf{*}}$	$\mathit{\sigma }$ or ${\mathit{\sigma }}^{\mathsf{*}}$
	bus	1.2	log-normal	0	0.12
accel	truck	1.3	log-normal	0.25	0.16
	PAC	2.6	log-normal	0.5	0.2
	bus	4	log-normal	−0.5	0.12
decel	truck	4	log-normal	−0.25	0.16
	PAC	4	log-normal	0.0	0.2
tau	all	1	log-normal	0.6	0.2
speedFactor	all	1	normal	1.0	0.1
sigmaerror	all	0.1	normal	0.1	0.01
ccoolness	all	0.95	normal	0.95	0.01
treaction	all	1	log-normal	−1.1	0.2

Table 1. Parameter distribution for cars.

Parameter	Vehicle Class	SUMO Default Value	Distribution	$\mathit{\mu }$ or ${\mathit{\mu }}^{\mathsf{*}}$	$\mathit{\sigma }$ or ${\mathit{\sigma }}^{\mathsf{*}}$
	bus	1.2	log-normal	0	0.12
accel	truck	1.3	log-normal	0.25	0.16
	PAC	2.6	log-normal	0.5	0.2
	bus	4	log-normal	−0.5	0.12
decel	truck	4	log-normal	−0.25	0.16
	PAC	4	log-normal	0.0	0.2
tau	all	1	log-normal	0.6	0.2
speedFactor	all	1	normal	1.0	0.1
sigmaerror	all	0.1	normal	0.1	0.01
ccoolness	all	0.95	normal	0.95	0.01
treaction	all	1	log-normal	−1.1	0.2

2.3.4. Driving Inaccuracy

To represent a real driver, the EIDM also provides the opportunity to adjust the inaccuracy of a driver. With the parameter speedfactor, the inaccuracy for exceeding or falling below the speed limit is represented. According to the results in [24], the selected speed is often around 2 km/h to 5 km/h above the actual speed limit if traffic density is low. As traffic density increases, the speed approaches the present speed limit. The distribution of speeds results in a normal distribution or a logarithmic normal distribution around the actual speed limit. Because urban situations mostly consist of higher traffic densities, the speedfactor is set lower than the mentioned values. For this reason, the work of [25] is used as a guide, as a validated vehicle model has already been developed here. The distribution used in this paper can be seen in Table 1.

Another parameter to define the drivers’ inaccuracy is the parameter sigmaerror, which describes the variance of the distribution, which effects the likelyhood and extent of the perception error, as explained in Section 2.2.

Because there are no literature values or studies from which this parameter can be derived, the simulation uses values drawn from a normal distribution mirrored around the SUMO default value, which also serves as the mean. The resulting parameter distribution is shown in Table 1. On the one hand, this means that there is no significant deviation from the default value in SUMO, as it can be assumed that it has a certain validity. On the other hand, a certain variance is nevertheless realized between different driver–vehicle units.

2.3.5. Further Adaptions and Validation

Additional to the already mentioned parameters, there are other parameters, which can be adapted. The most significant is the so-called coolness parameter. This parameter only has an indirect impact on the longitudinal motion of the vehicles but controls the lane change and threading behavior of vehicles. A value of 0 represents egoistic behavior and 1 represents cooperative behavior. Additionally, for this parameter, there are no values found in the literature or present studies, which would have provided a well-founded approach. Therefore, comparable with the approach for sigmaerror, a normal distribution around the SUMO default value is used, which is also mirrored as the maximum possible value of one.

To model buses and trucks, the distributions selected for passenger vehicles are adapted. Therefore, acceleration and deceleration are adjusted. Buses are modeled with the lowest values, as they have to fulfill a safety aspect in addition to their high weight. The transportation of passengers, some of whom are standing, means that only low acceleration rates are reasonable. At the same time, frequent halting at bus stops means that bus drivers have a high degree of repeat accuracy in terms of acceleration and deceleration. For this reason, the variance of the distributions is lower than for the distributions of passenger cars. Trucks represent an intermediate link between cars and buses. The expected value and the variance of the distribution therefore lies between those of buses and cars. The remaining parameters are aligned with those of passenger cars, as they primarily reflect human driving behavior, which is largely independent of vehicle class.

2.4. Modeling of Traffic Demand

By using the now-defined driver–vehicle unit parametrization, the traffic demand can be built up. To do so, there are already different methods mentioned in the literature. For example, in [26], a traffic simulation of the city of Berlin, Germany is built. As a reference, traffic counting data from 2014 was used. The focus of this work was modeling the main roads accurately by calculating the daily average utilization. Less-frequented roads were not focused on in this work.

Another example of building the traffic demand of a SUMO simulation is the Luxembourg SUMO traffic scenario [27], which uses demographic information about, e.g., number of inhabitants, age distribution and others to create urban traffic scenarios.

In [28], a SUMO traffic simulation of Monaco was built. In this simulation, the city was split into different zones. Traffic movement is generated based on the movement of vehicles between defined zones and the weighting of individual zones in terms of size and attractiveness.

These examples show that generating the traffic demand for a SUMO simulation can be very different and needs to be based on the individual characteristics of the area to be simulated.

Therefore, an approach tailored to the characteristics of the city center of Darmstadt was developed for this work. The traffic of Darmstadt is basically defined by a high percentage of commuter traffic, because many workplaces are located outside the city center and because the main streets of the city center serve as thoroughfares from the neighboring highway to the surrounding suburban areas.

To implement this in a SUMO simulation, the commuter traffic and the inner-city traffic are modeled separately, with a high focus on the commuter traffic. For the inner-city traffic RandomTrips, a tool provided within the SUMO framework is used to create random traffic with random driver–vehicle units within the street network.

This is supplemented by the creation of the commuter traffic. For this, traffic routes through Darmstadt are generated by using primarily main roads. For commuter traffic, a distinction is made between through traffic, traffic from the city center going away from it, and traffic coming from outside the city center going toward it.

In the first step, various routes have to be generated. Therefore, initial start and end points for the respective commuter traffic need to be set. For the through traffic, these points are set primarily on main roads at the edge of the simulation area. For traffic starting or ending in the city, random points within the city are connected with points at the edge of the simulation area, and also primarily on main roads. To generate routes based on these start and end points, they are connected by primarily using main roads as much as possible, because this represents general commuter traffic.

In the second step, these routes need to be assembled together over the simulation time to create a realistic traffic situation. Therefore, real traffic data, traffic count data from 2024, is used, which is provided from the city of Darmstadt [29]. This data is computed at special intersections and recorded as a sum of vehicles for each minute. The chosen intersections are shown in Figure 3. These are intersections with high traffic volumes as well as hubs for mobility in Darmstadt. For the traffic data, just work days (Monday to Friday) are used, as these account for the majority of traffic, and as events and activities on weekends can cause anomalies that could influence the representativeness of the data.

With this real traffic data, a heuristic, iterative gradient optimization is executed to assemble the predefined routes for steady traffic behavior for approximately over 1 h.

The bus demand of Darmstadt was build based on the actual bus timetable and the intervals between trips on the individual bus lines contained therein.

The demand of trucks is built by selecting a certain amount of vehicles from commuter traffic and the inner-city traffic to be executed as trucks. The amount is therefore selected by the proportion of trucks, corresponding to the proportion of registered trucks and cars in Germany [30].

2.5. Validation of the Traffic Simulation

To validate the driver–vehicle unit parameters and the traffic demand, real data was used for this purpose. The real driving data was obtained using a data logger that recorded key signals from the CAN bus of a Passat GTE, which is operated by the institute. The displayed data set just contains data from the given simulation area. Data was recorded over six years from 2018 to 2024. For the traffic data, the data mentioned in Section 2.4 from 2024 is used [29].

2.5.1. Vehicle Parametrization

To validate the longitudinal motion of the driver–vehicle units for all of the units in the simulation, all acceleration and velocity values are taken into account. This data is visualized in Figure 4a and can be compared to the real driving data in Figure 4b.

Even though the distribution and the progression of the quantiles differ slightly, the parametrization of the model can be assumed to be sufficient. The most likely reason for this is a very one-sided driver profile, because mostly scientists with an affinity for vehicles drove the Passat GTE. This does not constitute a fully representative group of drivers. Additionally, the fact that only one specific vehicle is driven makes comparison more difficult. Nevertheless, due to the large amount of data, a high degree of significance can be derived, and the corresponding correlation can be recorded as a validity feature for this use case.

2.5.2. Traffic Demand

The number of vehicles passing through the relevant intersections is adjusted as closely as possible to the actual traffic count data. The results for generating the traffic scenarios with the presented method are displayed in Figure 5. To compare the simulations with real traffic data, for the real traffic data, the mean value over each day is used. For the average traffic scenario, the data from the whole day is used. For the rush-hour traffic, just data between 7 a.m. and 10 a.m. and between 5 p.m. and 8 p.m. are included. The error bars represent the standard deviation of the average daily values over the whole of 2024. The reasons why an absolute match between the vehicle numbers is not possible are diverse, but are mainly due to the rule-based execution of vehicles in SUMO. For example, right-of-way rules are strictly followed. An agreement between several drivers to selectively disregard traffic rules in order to improve traffic flow is not possible.

Even though SUMO presents some limitation, the generated simulations show sufficient agreement with real traffic data. The two simulations can therefore be seen as representative traffic scenarios for the explained conditions. With this in place, they can serve as a basis for a consumption analysis of the vehicle fleet contained therein. The resulting findings can therefore be considered representative for Darmstadt.

3. Vehicle Consumption Calculation

To calculate the driving consumption of the vehicle fleet, a backward-facing longitudinal dynamics model is used. Therefore, at first, the driving resistances are calculated, and subsequently, the powertrain losses are included. For this work, losses of transmission, the battery system and electric machines (EMs) with inverters are included because they account for almost all of the driving consumption. In addition to the driving consumption, a variety of additional electrical consumers, known as secondary consumers, are connected to the on-board power system in electric vehicles. These secondary consumers do not directly contribute to traction power but support driving operations and, among other things, provide comfort and driver assistance functions. These secondary consumers are included in this work, which account for a significant proportion of total energy demand, depending on travel time and the environmental temperature $T_{\mathrm{out}}$. A schematic representation of the calculation process is shown in Figure 6. Here, $F_{\mathrm{res}}$ are the driving resistance forces, $P_{\mathrm{dri}}$ is the driving power, $T_{\mathrm{out}}$ is the outside temperature, and $P_{\mathrm{ges}}$ is the total consumption while driving.

3.1. Fleet Composition

For representative results of a fleet consumption analysis, the composition of the vehicle fleet is significant. The distribution of the three vehicle classes is set by the distribution within the SUMO simulation, as shown in Figure 1. In the average traffic scenario, there are $93.6\%$ PAC, $4.2\%$ trucks and $2.1\%$ buses included. For the peak-hour scenario, there are $94.9\%$ of the vehicles in the simulation classified as PAC, $3.8\%$ are trucks and $1.2\%$ are buses. The lower number of trucks and buses in the peak-hour scenario is reasonable because they are not affected by peak-hour traffic to the same extent as PAC.

Within the three considered vehicle classes, the vehicle characteristics can vary significantly. Therefore, for PAC and for trucks, another distinction needs to be made. Based on the “Mobility in Germany 2023” study [6], in a medium-sized city like Darmstadt, the share of small, compact and middle class vehicles is together over $90\%$. To represent this in the following modeling, three substitute vehicles are modeled which represent their vehicle class as accurately as possible. For the substitute vehicles, the VW eUp, representing segment A, the VW ID.3, representing segment C and the Audi Q4 E-TRON, which represents segment J, are chosen. The vehicle parameters are shown in Table 2.

Taking a look at the vehicle registration figures in Germany in 2024 [31], the share of small and microcars account for $24.9\%$ of the total PAC fleet in Germany, and the proportion of SUVs, off-road vehicles, and luxury and upper mid-range vehicles is $22.8\%$. On this basis, a distribution is chosen—$25\%$ of vehicles that use the VW eUp configuration, $25\%$ that use the Audi Q4 E-TRON configuration, and the remaining $50\%$ that use the VW ID.3 configuration—for further consideration in this paper.

For the truck fleet, a distinction is made between trucks of class N1 and class N2. N3 trucks are excluded because these heavy-duty trucks are primarily used for long-distance transport and are, therefore, not relevant for urban traffic. Additionally, there is a transit prohibition for heavy-duty trucks in Darmstadt. Based on the Federal Motor Transport Authority [32], by excluding N3 trucks, the resulting share is $85\%$ N1 trucks and $15\%$ N2 trucks. For this paper, N1 trucks are represented by the Maxus eDeliver 9 and the N2 trucks are represented by the MAN eTGL. The vehicle parameters are shown in Table 2.

Because of the relative low share of buses, no distinction is made within the bus fleet and a Mercedes eCitaro G is used as the substitute vehicle for all buses in the simulation. Also, the parameters are presented in Table 2.

Table 2. Parameter distribution for cars.

	Segment A	Segment C	Segment J	Segment N1	Segment N2	Bus
Deputy Vehicle	VW eUp	VW ID.3	Audi Q4 E-TRON	Maxus eDeliver 9	MAN eTGL	Mercedes eCitaro G
${\mathit{c}}_{\mathbf{w}}$	0.31	0.27	0.28	0.32	0.5	0.6
${\mathit{A}}_{\mathbf{v}}$ in ${\mathbf{m}}^{\mathbf{2}}$	2.08	2.3	2.56	4.2	6	8
${\mathit{m}}_{\mathbf{empty}}$ in kg	1235	1815	2235	2460	5400	20,365
${\mathit{m}}_{\mathbf{loading}}$ in kg	-	-	-	250	3300	5475
engine type	PMSM	PMSM	ASM (front) PMSM (rear)	PMSM	PMSM	ASM
max. Power in kW	60	150	70 (front) 210 (rear)	150	210	4 × 108.59
max. Torque in Nm	210	310	160 (front) 545 (rear)	310	750	4 × 485
gear ratio	8.16	11.53	9.95 (front) 7.915 (rear)	13.00	first: 20.86 second: 8.69	22.66
tire radius in m	0.29	0.34	0.37	0.36	0.38	0.48
sources	[33,34,35,36,37]	[38,39]	[40,41,42,43]	[44,45,46]	[46,47,48]	[46,49,50,51]

Table 2. Parameter distribution for cars.

	Segment A	Segment C	Segment J	Segment N1	Segment N2	Bus
Deputy Vehicle	VW eUp	VW ID.3	Audi Q4 E-TRON	Maxus eDeliver 9	MAN eTGL	Mercedes eCitaro G
${\mathit{c}}_{\mathbf{w}}$	0.31	0.27	0.28	0.32	0.5	0.6
${\mathit{A}}_{\mathbf{v}}$ in ${\mathbf{m}}^{\mathbf{2}}$	2.08	2.3	2.56	4.2	6	8
${\mathit{m}}_{\mathbf{empty}}$ in kg	1235	1815	2235	2460	5400	20,365
${\mathit{m}}_{\mathbf{loading}}$ in kg	-	-	-	250	3300	5475
engine type	PMSM	PMSM	ASM (front) PMSM (rear)	PMSM	PMSM	ASM
max. Power in kW	60	150	70 (front) 210 (rear)	150	210	4 × 108.59
max. Torque in Nm	210	310	160 (front) 545 (rear)	310	750	4 × 485
gear ratio	8.16	11.53	9.95 (front) 7.915 (rear)	13.00	first: 20.86 second: 8.69	22.66
tire radius in m	0.29	0.34	0.37	0.36	0.38	0.48
sources	[33,34,35,36,37]	[38,39]	[40,41,42,43]	[44,45,46]	[46,47,48]	[46,49,50,51]

3.2. Driving Consumption

The driving resistance forces $F_{\mathrm{dr}}$ consist of acceleration resistances $F_{\mathrm{ar}}$, aerodynamic drag $F_{\mathrm{ad}}$, rolling resistances $F_{\mathrm{rr}}$ and incline resistances $F_{\mathrm{ir}}$ and are calculated based on Equation (6) [52]. As a result, the required traction torque can be calculated with the dynamic tire radius. The vehicle-specific parameters for this purpose are provided in Table 2.

$$\begin{array}{cc}\hfill F_{\mathrm{dr}}& =F_{\mathrm{ar}}+F_{\mathrm{ad}}+F_{\mathrm{rr}}+F_{\mathrm{ir}}\hfill \\ \hfill & =m_{\mathrm{eff}}a+{\displaystyle \frac{1}{2}}{c}_{\mathrm{w}}{A}_{\mathrm{v}}\rho v^2+c_{\mathrm{r}}{m}_{\mathrm{eff}}gcos{\alpha }_{\mathrm{s}}+m_{\mathrm{eff}}gsin{\alpha }_{\mathrm{s}}\hfill \end{array}$$

(6)

In this equation, the rotational inertia of the drivetrain is neglected for simplification, so that $m_{\mathrm{eff}}=m_{\mathrm{empty}}+m_{\mathrm{loading}}$. Here, $m_{\mathrm{empty}}$ is already calculated with one driver with 75 kg. The variable $\rho$ denotes the air density, while ${\alpha }_s$ represents the current slope in the driving direction. The rolling resistance coefficient is used as $c_r=0.0075$ and the gravity is given with g. The values for the frontal surface $A_v$ and drag coefficient $c_w$ can be seen in Table 2.

With the required torque calculated, the losses in the powertrain can be added. Constant efficiency factors are assumed for the battery and transmission. Since the efficiency of transmissions varies significantly less than the efficiency of motors and power electronics depending on the load point, according to [25], this paper assumes a constant transmission efficiency regardless of the load point.

The battery efficiency is assumed to be $95\%$, based on [53]. For the fixed gear transmission, an efficiency of $98\%$ is adopted, following [46]. In the case of the MAN eTGL, which is equipped with a shiftable two-speed transmission, an efficiency of $95\%$ is assumed. This lower value reflects the assumption that increased mechanical complexity typically leads to higher transmission losses.

For the EM and inverter combination, characteristic efficiency maps are used. These are taken from the literature and are either identical to the EMs used in the vehicles (VW eUp [37], VW ID.3 [54], AUDI Q4 E-Tron (rear axel) [55]) or equal to the EM design, power and torque (AUDI Q4 E-Tron (front axel) [43], Maxus eDeliver9 [54], MAN eTGL [56]) or scaled according to [57] to get a representative efficiency map (Mercedes eCitaro G [58]). If the efficiency map in the source does not include inverter efficiencies (only relevant for [54]) a constant inverter efficiency of $97\%$ is added, which is based on [59].

With this in place, the total driving power $P_{\mathrm{dri}}$ can be calculated, which is supplemented by secondary consumers.

To validate these powertrain simulations with the given parameters, the consumption of the PAC based on the WLTP and the WLTP-Low are compared to the simulated vehicles. The results can be seen in

Table 3. Comparison of simulated driving consumption and real consumption analysis.

	Segment A	Segment C	Segment J	Segment N1
Deputy Vehicle	VW eUp	VW ID.3	Audi Q4 E-TRON	Maxus eDeliver 9
real consumption WLTP in kWh/100 km	14.5 [60]	15.3 [61]	16.8 [40]	26.2 [62]
simulated consumption WLTP in kWh/100 km	12.5	14.8	16.9	25.4
deviation simulations vs. real in %	13.8	3.2	0.5	3.0
real consumption WLTP-Low in kWh/100 km	10.0 [63]	11.5 [64]	12.5 [40]	16.6 [62]
simulated consumption WLTP-Low in kWh/100 km	8.5	11.6	13.4	17.4
deviation simulations vs. real in %	15.0	0.8	6.7	4.5

.

Except for the VW eUp, all other vehicles show a deviation of less than $5\%$, which is a high accuracy based on the fact that efficiency maps are at some points adjusted and operating strategies are unknown. The 10–$12\%$ deviation for the eUp probably result from inaccuracies within the efficiency map in [37] or less energy can be recuperated, than assumed in this work. This has to be taken into account when discussing the following results regarding fleet consumption.

For the MAN eTGL and Mercedes eCitaro G, no official consumption analysis can be found. In [65], for a Mercedes eCitaro G, consumption including secondary consumers of 150 kWh/100 km is measured for buses within Darmstadt. No distinction is made here according to type of consumption, but the results in Section 4 suggest that the assumptions in this work are close to the actual consumption. For these two vehicles, the validity of the simulations can be derived from the methodology, which, as shown, leads to a high degree of accuracy for other vehicles.

3.3. Secondary Consumption

Especially for the consumption of BEVs, it is important to model secondary consumers of the vehicles because of the higher influence of secondary consumers on consumption in comparison to internal combustion vehicles [7]. For BEVs, the secondary consumers do not influence the operation point of the motor and therefore can be modeled as an additional energy extraction out of the battery [2]. In general, the secondary consumers can be divided into temperature-independent and temperature-dependent, the power demand of the secondary consumers is not time-dependent, because pre-conditioning of the vehicles is not considered in this work. As follows, the energy demand scales linear with driving time.

The temperature-independent secondary consumers of the vehicles are subdivided into four different categories. In the category lighting, interior (e.g., reading light) and exterior (e.g., daytime running lights, direction indicator lights) lighting is considered. Lighting is responsible for three to four percent of the secondary consumers’ energy demand. The comfort, infotainment and connectivity category constitutes between $4\%$ and $7\%$ of the secondary consumers energy demand, e.g., for the audio system and on-board computer. The category of vehicle systems and safety, consisting of, e.g., windscreen wipers and sensors, leads to between $13\%$ and $17\%$ of the energy demand of secondary consumers. Control units and pumps are part of the category of driving functions and constitute between $11\%$ and $29\%$ of the secondary consumers’ energy demand. A detailed line-up of the energy demand of temperature-independent secondary consumers including relevant literature can be found in Appendix A, and the summed-up values for the used vehicles in this work are shown in

Table 4. Secondary consumer power for different vehicles.

Deputy Vehicle	VW eUp	VW ID.3	Audi Q4 E-TRON	Maxus eDeliver 9	MAN eTGL	Mercedes eCitaro G
Temperature-independent secondary consumer power [W]:	455.0	517.8	522.8	517.8	817.8	3506.0
Temperature-dependent secondary consumer power [W]:	806.2	870.1	970.8	806.2	806.2	7711.0
Total secondary consumer power [W]:	1261.2	1387.9	1493.6	1324.0	1624.0	110,217.0

.

The temperature-dependent secondary consumer of the vehicles is the Heating, Ventilation, and Air conditioning (HVAC) system. The energy demand for conditioning the passenger compartment is dependent on the inside and outside temperature of the vehicle. For the inside, a temperature of 22 °C is chosen based on [66]. For the outside, the mean temperature of Darmstadt between 6 am and 10 pm is chosen to 11.78 °C, because no driving during the night is being considered [67]. To calculate the stationary energy demand for conditioning the passenger department, a simplified thermodynamic model is used, considering air mass flows, heating through passengers and solar radiation and conduction through the bodywork of the vehicle. The summed-up energy demands of temperature-dependent secondary consumers for the used vehicles, as well as the total energy demands of secondary consumers, are shown in Table 4.

4. Results

In this section, the average consumption of the vehicle fleet within the traffic environment is presented. Therefore, the results for average traffic and for peak-hour traffic are presented. At first, the driving consumptions are investigated for each vehicle and based on the composition for each vehicle class. Following this, the results including secondary consumers are presented for each vehicle and vehicle group. With these results and the fleet composition, the fleet consumption for the two traffic scenarios are presented.

4.1. Driving Consumption

The average consumption is calculated based on the methodology in Section 3.2. For each PAC, the results and the specific proportions can be seen in Figure 7.

Here, an increase in consumption for heavier and bigger cars can be proven, as expected with higher vehicle weights. Also, the consumption does not differ significantly between mean traffic and peak-hour traffic. The negative incline resistance for these vehicles is based on the slope within Darmstadt and a higher number of vehicles going downwards than upwards. The biggest differences between mean and high volume traffic is based on rolling resistances, aerodynamic drag and EM losses. This is based on the lower speed within the peak-hour traffic scenario (average speed of the peak-hour traffic scenario is $13.2$ km/h and $23.3$ km/h for the average traffic scenario), which results in driving within areas of lower efficiency of the EMs. Furthermore, it reduces aerodynamic drag and increases the rolling resistance.

The comparison between average traffic and peak-hour traffic shows no big deviation. This can be attributed to the fact that the potentially lower efficiency of electric motors at reduced speeds is compensated by the generally lower driving demands at these speeds, such as reduced acceleration.

The driving consumption for the deputy trucks, the truck fleet and the bus can be seen in Figure 8.

The results for trucks and the buses do not differ significantly from the PAC results. Also here, the total consumption primarily differs between the vehicles based on weight. As mentioned, for the MAN eTGL, the two-gear transmission demands higher losses due to shifting losses, compared to the other mentioned one-gear transmissions.

The bus consumption is caused by its weight and shape, as it is the vehicle with the highest individual consumption.

4.2. Secondary Consumer

Additionally to the average driving consumptions, the average SC needs to be added to the driving consumption to get an accurate picture of the energy consumption of a vehicle fleet. With the methodology of Section 3.3, the following results can be obtained. In Figure 9, the consumption of passenger vehicles, divided into the SC and driving consumption, can be seen.

With this, the influence on vehicle consumption and higher traffic volume becomes clear. Even though the difference between the driving consumption of the vehicles of the two traffic scenarios is very similar, the SC consumption differs significantly in this regard. This is caused by the lower mean speed and therefore higher travel time in this scenario. Because the SC is based on time and not on distance traveled, this has a significant impact and can be verified for all vehicle classes.

The majority of the SC is based on the temperature-dependent SC, which is dependent on travel time and weather condition, so this can also be higher within colder weather scenarios.

The SC is also analyzed for the truck and bus fleet. The results can be seen in Figure 10.

For the truck fleet, the SCs are in absolute values similar to the PAC consumptions. This is based on the similar passenger cabin, which needs to be conditioned, and the similar temperature-independent secondary consumers. Based on the higher driving consumption, this accounts for a smaller proportion of total consumption for these vehicles.

For the bus fleet, the SC is greatly disproportionate, compared to the driving consumption. This is based on the large interior PAC that needs to be temperature controlled. Typically, oil heaters are used in these buses to create a comfortable indoor temperature to reduce driving consumption.

For the sake of comparability, we also include this energy demand—fully electric heating is assumed here—which results in the consumption presented in Figure 10b.

4.3. Fleet Consumption

With the previous results in place, the average consumption of the whole fleet can be calculated. With the mentioned marked shares of the different vehicles and vehicle classes, the results in Figure 11 can be obtained.

Within these results, it becomes clear that the higher consumption of trucks and buses is offset by the high number of passenger vehicles. Therefore, the average consumption of the vehicle fleet is relatively close to the PAC consumption. But with the comparison of peak-hour and average traffic scenario, the impact of the vehicle classes can be seen. The 1 kWh/100 km lower consumption in the peak-hour traffic scenario is therefore based on the slightly lower share of buses and trucks in this scenario, which show greater individual consumption.

Looking at the SC, it can be stated that there is a major impact on urban vehicle fleet consumption, being $36\%$ of the total energy consumption in the average traffic scenario and $48\%$ in the high-traffic scenario. The impact of the SC results in the highest energy consumption for the peak-hour traffic scenario, even though a lower average driving consumption can be obtained.

In order to compare the results with current traffic, Figure 12 compares the average fuel consumption and $C{O}_2$ emissions of a BEC fleet with those of a vehicle fleet based purely on combustion technology. The all combustion vehicle fleet is based on urban consumption data for each vehicle class.

In [68], vehicle consumption data was analyzed for driving in urban area in Riobamba, Ecuador. The results show a consumption of $13.2$ km/L to $13.9$ km/L, which is the equivalent of $7.2$ L/100 km to $7.6$ L/100km, with the density of gasoline and the calorific value, the consumption, in kWh/100 km, can be calculated. This consumption is applied for all passenger vehicles.

For trucks, the study by Zamboni et al. [69] was used. Here, combustion trucks were investigated for their consumption in low speed urban and port areas. The study estimates consumption in urban areas at $0.3$ kg/km, which is equivalent to $35.9$ L/100 km diesel. Here, the density and heat value of diesel are used to calculated the consumption of the truck fleet.

For the buses, the studies of Zhang et al. [70] and Keramydas et al. [71] were considered. Here, buses in urban areas were investigated based on their fuel consumption. In [70], the results show a consumption of around 40 L/100 km. In [71], the consumption is stated in a range of 40 kg/100 km to 60 kg/100 km, which is equivalent to $47.8$ L/100 km to $71.9$ L/100 km. The higher consumption for the investigations in [71] are resulted by significantly higher average speeds. Assuming that the bus speed in this study lies between the average speeds reported in the two referenced investigations, the fuel consumption for buses is estimated at 50 L/100 km, based on the cited sources. Furthermore, here, the density and heat value of diesel are used to calculate the consumption of the bus fleet.

The results obtained can be seen in Figure 12a. The low efficiency of combustion engines compared to electric drivetrains is very clear here and results in a more-than-double energy consumption for a pure combustion engine vehicle fleet.

Taking the electricity mix in Germany between October 2023 and 2024 [3], the $C{O}_2$ emissions based on the average consumption can be calculated. This can be seen in Figure 12b. Because the conversion from electric energy consumption to $C{O}_2$ emission is directly proportional, the equivalent difference can also be seen here.

4.4. Discussion

The presented results provide insights into how different traffic conditions and vehicle parameters influence the energy consumption of pure BEV fleet in urban settings. The simulation framework isolates the key factors and their quantifiable impact on overall consumption.

The shown results are giving a trend, how the average energy consumption of BEVs in an urban environment is composed and which are the main influences. All influencing factors are based on the literature and are largely validated. Even though the results are significantly based on these factors. For changes regarding fleet composition, vehicle parameters or traffic behavior, these results can change significantly.

Also in this paper, the average consumption for all vehicles are calculated. This value is averaged over 100 km to also compare the different traffic scenarios. The absolute energy demand and $C{O}_2$ emission is therefore a result from traffic volume and this consumption. Therefore, even though the average consumption between the scenarios just differs by around 4 kWh/100 km, the total energy demand will be significantly higher due to the significantly higher number of vehicles.

5. Conclusions

In this paper, a method to build and validate a traffic simulation using SUMO with traffic and vehicle data is proposed. Additionally, different powertrain models were built and validated based on accessible consumption data. In order to obtain realistic fleet consumption figures, the SC was modeled.

With this, the average consumption for a BEV vehicle fleet in high dense traffic and in average traffic was modeled for Darmstadt, Germany. For these scenarios, the average $C{O}_2$ emission while driving is significantly lower for all BEV fleets compared to combustion technology fleets. Even if these results support the rationale for EU legislation in regard of $C{O}_2$ emissions in the investigated urban areas [1], only looking at emissions during driving, though the emissions are analyzed beyond local emissions, is not sufficient. A Life-Cycle Assessment, including the production and recycling phases in addition to the use phase, needs to be conducted to elicit a comprehensive assessment of total $C{O}_2$ emissions and evaluate the suitability of a vehicle for minimal overall lifetime $C{O}_2$ emission, as well as looking at representative driving scenarios including, e.g., highway driving. Further factors like financial, social or energy grid concerns are not included in the presented investigation.

Nevertheless, these results can be progressively used to perform detailed analysis of the vehicle consumption, for example, to investigate energy-saving potential.

The results show that regarding driving consumption, the main aspects to consider are the rolling resistance and the losses caused by the EMs. In order to reduce vehicle consumption, this suggests that, in the future, attempts should be made to reduce losses in the drivetrain, for example, by using drivetrains specially adapted for their application [72].

It is also clear that the influence of secondary consumers is affected by lower speeds, due to time dependency. This suggests that in order to save energy especially for future all BEV fleets, the research field of traffic coordination, e.g., by traffic light coordination gets even more important in order to increase traffic flow and average velocity. These traffic management technologies should be further developed so that they prevent congestion even more effectively and maintain traffic flow as best as possible.

Looking into the future, with vehicles containing advanced driver assistance systems or autonomous vehicles in urban environments, maximizing vehicle speeds within the permitted speed limits should also be a goal in order to save energy, and consequently $C{O}_2$, in addition to safety aspects.