Exploring cost performance tradeoffs and uncertainties for electric- and autonomous electric trucks using computational experiments

Abstract

The recent development of battery electric trucks (BETs) suggests that they could play a vital role in transitioning to zero-emission road freight. To facilitate this transition, it is important to understand under which conditions BETs can be a viable alternative to internal combustion engine trucks (ICETs). Concurrently, the advancement of autonomous driving technology adds uncertainty and complexity to analyzing how the cost competitiveness of future zero-emissions trucks, such as autonomous electric trucks (AETs) may develop. This study examines the cost performance of BETs and AETs compared to ICETs, and how it varies over different market and technology conditions, charging strategies, and transport applications. Focus is on heavy-duty tractor-trailer trucks operating full truckload shuttle-flows in Sweden. Due to the inherent uncertainty and interactions among the analyzed factors, the analysis is performed as computational experiments using a simulation model of BET, AET, and ICET shuttle flow operations and associated costs. In total, 19,200 experiments are performed by sampling the model across 1200 scenarios representing various transport applications and technical and economic conditions for sixteen charging strategies with different combinations of depot, destination, and en route charging. The results indicate that both BETs and AETs are cost competitive compared to ICETs in a large share of scenarios. High asset utilization is important for offsetting additional investment costs in vehicles and chargers, highlighting the importance of deploying these vehicles in applications that enable high productivity. The cost performance for BETs is primarily influenced by energy related costs, charging strategy, and charging infrastructure utilization. The AET cost performance is in addition heavily affected by remote operations cost, and costs for the automated driving system. When feasible, relying only on depot charging is in many scenarios the most cost-effective charging strategy, with the primary exceptions being highly energy-demanding scenarios with long distances and heavy goods in which the required battery is too heavy to operate the truck within vehicle weight regulations if not complemented by destination, or en route charging. However, many experiments do not lead to a reduced payload capacity for BETs and AETs compared to ICETs, and a large majority of the considered scenarios are feasible to operate with a BET or AET within current gross vehicle weight regulations.

1 Introduction

With the anticipated growth in freight demand, it is imperative transition to low- or zero carbon energy sources to decarbonize road freight transport [14]. Recent research and market trends suggest that battery electric heavy-duty trucks (BETs) can play a key role for this decarbonization [2, 9, 26]. A key condition for a large-scale adoption of BETs is that they are cost competitive with conventional internal combustion engine trucks (ICETs) and other alternative fuel technologies for a wide range of transport applications. It is therefore important to understand the circumstances and transport applications where BETs can be a cost-competitive and technically viable alternative, and how this may develop over time. Concurrently, the advancement of autonomous driving technology holds promise to improve the cost performance of BETs, but its development trajectory is uncertain which further complicates the analysis of future zero-emissions trucks, notably autonomous electric trucks (AETs).

There is a growing literature on BET cost performance. Less work has been done on autonomous trucks, and this literature has primarily studied autonomous ICETs, and not AETs. The current literature suggests that the outcomes of BET and AET cost performance assessments depend on three interdependent domains: 1) the specifics of the transport application the BET/AET is intended for; 2) the current and future market and technology conditions; 3) the design of the BET and AET solutions, including charging strategy and battery sizing.

On a high-level, the cost performance of BETs is about balancing their lower variable operating costs against the higher initial investments required for battery and charging infrastructure [29]. With AETs this likely becomes more pronounced when labor-related costs are reduced, but technology costs increase [31]. Reaching a breakeven point requires a certain amount of truck productivity, e.g. in terms of transportation work performed, which is contingent upon the transport application [28]. Generally, a high utilization of the vehicle, battery, and charging infrastructure is important to lower the fixed costs per unit of transport work to a competitive point [8, 11, 15].

Cost performance analyses of BETs and AETs are sensitive to assumptions and uncertainty in techno-economic parameters such as the price of batteries, energy and automated driving systems (ADS) [8, 11, 21, 23, 28]. Also, future policies and incentives for electric vehicles influence the cost competitiveness [2, 3]. Furthermore, the availability of charging infrastructure affects which transport routes are cost efficient to electrify and what battery size and charging strategy are cost optimal [27]. For AETs there are additional considerations, for instance that they may require support from remote human operators for which the costs and performance is yet highly uncertain [8, 11, 32].

It is important to carefully design BET or AET solutions for the transport application they will serve. A key consideration is the interdependent choice of battery capacity and charging strategy [16, 24]. If charging is primarily done between shifts, a heavy battery may be required which may impose a payload constraint [20, 21]. High-power intermittent charging can reduce the required battery size but also leads to increased charging costs [16], unproductive standstill time [21] and intensified degradation of batteries [12]. Fleet-level studies of real world transport networks have highlighted additional considerations such as the fleet composition, routing, logistical constraints and available time for charging [4, 10, 19]. AETs introduce further opportunities to optimize the operations, such as adjusting driving speeds and routing [5, 6].

This paper demonstrates a model-driven approach for analyzing how the cost performance of BETs and AETs is determined by three interdependent domains: the transport application, market and technology conditions, and charging strategy. The analysis approach combines tools for multivariate computational experiments with truck-level techno-economic modeling of BET and AET operations. The model accounts for key factors such as energy constraints, battery dimensioning and degradation, charging infrastructure dimensioning and costs, and AET related costs. The experiments are performed by sampling the model over a parameter space spanning these three domains. The main motivation for this exploratory modeling approach is that technoeconomic analyses of electric trucks are sensitive to the specifics of the studied transport application and assumptions on technoeconomic parameters, often with substantial uncertainty about how they will develop in the future.

The results of the experiments are then analyzed to assess how the cost performance is affected by the varied factors, and to identify regions in the parameter space that have high concentrations of cost competitive BET or AET cases which provides an indication of boundary conditions for cost effective and cost robust BET and AET deployment.

The approach is demonstrated for medium- to long-distance shuttle flows with full truckloads in a Swedish setting. While this is a rather simple type of transport application it serves as a useful case for showcasing the approach and it has practical relevance since it is a likely initial application for BETs and AETs in the short- to midterm as it enables a high battery utilization [16], access to charging [4] and often exhibits a high degree of repetitiveness and low operational complexity which is favorable for AETs [18, 22]. This paper is, to the best knowledge of the authors, only the second study comparing cost performance between ICETs, BETs and AETs, second to Ghandriz et al. [11]. Specific additional contributions from this paper are the inclusion of en route charging, comparison of different charging strategies, impacts from multi-shift operations and methods for mapping the high-dimensional input space to the outcomes.

2 Methods

A series of computational experiments is performed to analyze the feasibility and cost performance of BET and AET for shuttle flows by sampling a model that simulates truck operations and associated costs of a BET, AET and ICET. Experiments are constructed by combining a set of scenarios representing different types of shuttle flows operated under different technological and market conditions, with a set of charging strategies. The remainder of this section outlines the considered transport application, provides a summary of the core model and the experimental setup, and defines the cost performance metrics. For details on the model, see supplementary materials part A, and for the experimental setup, see supplementary materials part B.

2.1 Transport application

The analyzed transport application is full truckload shuttle-flows between a single origin and single destination that can be completed during one or two driver shifts. The shuttle is operated by a tractor-trailer truck which pulls a loaded trailer from the origin to the destination, where it is dropped off. An empty trailer is then picked up and returned to the origin. Focus is on medium- to long-haul shuttles of 100–550 km one-way driving in Sweden. In 2022, 56% of the total driven truck-kms by Swedish trucks was generated by trips of 100–500 km, and trips longer than 500km accounted for an additional 20% [30]. It is assumed that only one trip is performed per shift and that the shuttle is operated repeatedly over the vehicle life. If the trip can be completed as a roundtrip within a single driver shift, the mandatory rest break is performed at the destination. If not, the break is assumed to take place in the middle between the origin and destination. Three types of charging are considered: depot charging at the origin between shifts, charging at the destination before the return trip, and en route charging performed halfway between the origin and destination.

2.2 Core model

The core model simulates truck operations and associated costs of a BET and AET over its vehicle life from a first buyer perspective and compares it to an ICET for each experiment. The model is implemented in Python. Scripts for setting up, running, and analyzing the experiments have been developed specifically for this study and make significant use of the open-source Python library EMA workbench (Kwakkel 2017) while the core model is proprietary and developed by the authors.

The calculation steps for one experiment are summarized in Fig. 1 and the remainder of this section. A detailed model description is provided in the supplementary materials (part A).

Fig. 1

Overview of the calculation steps and parameters for an instance of the core model. For a detailed model description with an in-depth description of each step, please see the supplementary materials (part A)

First, a vehicle schedule is generated. This is a simplistic representation of how the truck is intended to be operated throughout its lifetime, shift by shift and day by day. For each shift, an energy simulation is performed by first calculating the energy consumption for each leg of the trip, and then accounting for any recharged energy from destination and en route charging, see Fig. 2.

Fig. 2

Schematic illustration of the energy modelling for different charging strategies. Negative slope indicates that the vehicle consumes energy (driving) and positive slope that the vehicle is charging. E_{en route} and E_destination denote the recharged energy during en route and destination charging respectively. The dimensioning energy need, E_dim, is the minimum net battery capacity required for the trip, E_depot, is the energy to recharge before the next trip

Battery sizing and dimensioning of depot charging power is performed endogenously depending on the results of the energy simulation. Since shuttle operations have predictable energy demands, it is assumed that the battery is dimensioned to provide sufficient range, with some margin for deviations around the nominal consumption, while maintaining the battery within a defined State-of-Charge (SoC) window. It is assumed that depot charging is used to fully charge the truck before commencing the next trip, and that the depot charging power is dimensioned to enable a full recharge between trips. The use of a power-dependent charging infrastructure cost model enables accounting for how the scenario and charging strategy impact the depot charging power and consequently overall cost performance. However, it is acknowledged that other principles for dimensioning depot charging infrastructure might be applied in real world operations. The vehicle weight calculation considers the tractor and trailer body, powertrain components, and the battery. The tractor and trailer body weight are assumed equal for ICE and BET but differ for the AET due to additional ADS weight and weight reductions if designed without a driver’s cabin. Energy consumption is calculated using a weight-dependent model of specific energy consumption (kWh/km).

A truck might not be able to complete the vehicle schedule, either due to payload restrictions (grossing out), by reaching its end of life in terms of total distance driven (distancing out) or due to battery degradation (cycling out). Battery degradation is modeled with focus on cycling degradation and the influence of charging C-rate.¹

Total life cycle costs are calculated for each truck type and include truck investment costs, variable operating costs, fixed operating costs, and residual value. Investment costs comprise truck and trailer chassis, powertrain, battery, and ADS. Variable costs are calculated per shift and cover energy, driver, maintenance, tire, and charging infrastructure. For the AET, also costs for remote operations (modeled as an hourly cost) and an additional service cost for preparing the vehicle for its next trip are included. Charging infrastructure cost is modeled as a user cost per unit of charged energy (kWh) and considers various investment and operational costs. The fixed operational costs cover insurance, vehicle tax and financing. The residual value is calculated separately for the truck and battery as fractions of their investment costs determined by the accumulated driven distance and battery degradation.

2.3 Experimental setup

The research design is inspired by Exploratory Modelling and Analysis [1, 17] and organized according to Fig. 3 by separating the considered parameters into three domains:

• Application domain (A): parameters specifying the characteristics for the shuttle flow.

• Market and technology domain (X): parameters specifying external factors such as electricity and fuel prices, battery performance and costs, ADS costs, and drivetrain efficiency.

• Charging strategy domain (L): parameters for the charging strategy.

Fig. 3

Overview of the structure and terminology for the experiment design

An experiment denotes a combination of a scenario (a point in the joint parameter space of the Application and Market and technology domains (A, X)), and a strategy (a point in the Charging strategy domain (L)). The core model (R) maps the experiment parameter space (A, X, L) to the cost performance metrics (M). Scenarios are generated using Latin hypercube sampling with uniform parameter distributions [13] while charging strategies are designed using a custom fractional factorial design based on discrete factors. The set of experiments is constructed as a full factorial over the scenarios and strategies which enables variations in feasibility and cost performance to be studied over the full set.

All experiments are performed for trucks with a 40-ton maximum gross vehicle weight (42 ton for BET and AET²). Key assumptions are: the battery capacity can be dimensioned without constraints, the AET is capable of operating the considered shuttle from a technical and regulatory perspective, and the specified charging infrastructure is available. The analysis is designed to represent general characteristics of generic BET and AET technology and not specific brands or vehicle models. The scenario parametrization (see Table 1 and supplementary materials part B) is intended to represent the following:

• ICET: a typical new semi-trailer tractor at the European market under current and near future conditions.

• BET: a new BET semi-trailer tractor at the European market. The BET parameters in the market and technology domain are used to represent different stages of technology maturity, with the less favorable end of the range representing current conditions and the full range plausible future improvements due to technological development.

• AET: the modeling is not based on specific designs or near-market models but is inspired by proposed solutions ranging from conventional trucks retrofitted with automated driving systems to more futuristic designs without a driver’s cabin. The AET analysis is exploratory and the AET-specific parameters in the market and technology domain represent an uncertainty range of plausible AET performance when the technology is mature enough to operate the considered type of shuttle flows.

Table 1 Scenario parameters and their ranges used for designing the experiments. Equations in the column Description show how the parameter is applied in the model (see supplementary materials part A for details on the model)

The study assumes a Swedish setting which is reflected in parameter settings, and the use of Swedish kronor (SEK). Each experiment is run for a period of up to ten years.

The scenario space is spanned by the 24 parameters summarized in Table 1. An in-depth description and motivation of these parameters is given in supplementary materials (part B) along with parameter values for the input parameters that are held constant between experiments.

Three charging strategy types (CSTs) are considered. These are separated by what type(s) of charging set up is used: CST1: depot charging only, CST2: depot and destination charging, CST3: depot, destination and en route charging. In total, sixteen distinct charging strategies are considered, each belonging to one CST, with different charging power used for destination and en route charging. Please see supplementary materials part B for a complete list of charging strategies. Currently, charging at 500kW or more is uncommon, but is expected to become more widely available in the coming years. To analyze the influence of availability of destination and en route charging above 500kW, CST2 and CST3 are split into two types: low power (denoted by suffix LP) for which both the destination and charging power < 600kW, and high power (denoted by suffix HP) for which either the destination or charging power ≥ 600kW.

Lever combinations considered for the design of charging strategies. CST2 and CST3 are further separated into low power (CST2/3LP) and high power (CST2/3HP).

	Charging strategy type (CST)
	CST1	CST2	CST3
Charging set up	{Depot}	{Depot, Destination}	{Depot, Destination, En route}
Charging power Destination${P}_{destination}$		{150, 300, 600} (kW)	{150, 300, 600} (kW)
Charging power en route${P}_{en route}$			{150, 300, 600, 1000} (kW)
# charging strategies	1	3	12

In total, 19,200 experiments are performed by combining the 16 charging strategies with the 1200 scenarios. Testing all charging strategies for each scenario without a-priori assumptions on charging strategy selection aligns with the approach of an exploratory model-driven analysis. However, this design implies that for each scenario, a different number of experiments per CST will be performed since the CSTs contain a different number of charging strategies. Therefore, much of the analysis is done on a scenario and CST level where the best performing charging strategy for the specific scenario is used to represent the CST. Another implication of the design is that all regions of the scenario space may not be equally likely to occur and therefore, the frequency of scenarios within a given scenario parameter region should not be interpreted as the probability for this type of scenario to occur.

Before the cost performance analysis, a feasibility screening is done using two criteria: i) the truck needs to be able to carry the payload without exceeding the maximum allowed GCW, and ii) the depot charging power should be less than 300kW since it is assumed that grid power access will be a limiting factor for most truck depots. Experiments not meeting both these criteria are classified as unfeasible and discarded from the subsequent cost performance analysis.

2.4 Cost performance metrics

Since the core model is run for all three truck types in each experiment, a direct comparison of the trucks’ relative cost performance within experiments can be made. The main metric used for analyzing cost performance is the relative difference in total costs normalized by the total number ton-km, for BET and AET compared to ICET, in experiment e. Cost performance is denoted as ${\rho C}_{tk{m}_{t=\left\{BET,AET\right\},e}}$ where t indicates truck type, and is defined according to EQ 1:

$$\begin{array}{c}{\rho C}_{tk{m}_{t,e}}=\frac{{C}_{tk{m}_{t,e}}-{C}_{tk{m}_{ICE,e}}}{{C}_{tk{m}_{ICE,e}}}\end{array}$$

(1)

where ${C}_{tk{m}_{t,e}}$ is total costs normalized per unit of transport work (tkm) for truck type $t$ and experiment $e$, as given by EQ 2:

$$\begin{array}{c}{C}_{tk{m}_{t,e}}=\frac{{C}_{tota{l}_{t,e}}}{{\sum }_{\in {i}_{t},e}{D}_{loade{d}_{i,t}}*{m}_{payloa{d}_{i,t}}}\end{array}$$

(2)

where ${C}_{tota{l}_{t,e}}$ is the total life cycle cost for the truck of type $t$, and ${\sum }_{\in {i}_{t},e}{D}_{loade{d}_{i,t}}*{m}_{payloa{d}_{i,t}}$ is the total transport work (tkm) performed over the set of shifts the truck operated, ${i}_{t}$. This formulation means that lower values for ${\rho C}_{tk{m}_{t,e}}$ correspond to better cost performance for BET or AET. The criterion for a BET or AET being cost competitive in an experiment is ${\rho C}_{tk{m}_{t=\left\{BET,AET\right\},e}}<0$ i.e. a lower cost per unit of transport work than the ICET.

3 Results

Many scenarios are feasible to operate with BETs or AETs, and in many scenarios at a lower cost per TKM compared to the ICET, see Fig. 4. The share of feasible scenarios increases with access to destination charging (CST2) and en route charging (CST3), but the use of high-power charging only has a minor effect on feasibility. For CST1, almost all feasible scenarios would also be cost competitive to operate with a BET or AET, while a somewhat smaller share of the scenarios feasible for CST2 and CST3 also are cost competitive. The remainder of this section details the BET and AET feasibility and cost performance across experiments.

Fig. 4

Feasibility and cost competitiveness for BET and AET over the 1200 scenarios separated by charging strategy type. Feasible denotes the number of scenarios where at least one strategy within the CST is feasible while Feasible and Competitive are scenarios where at least one strategy is both feasible and cost competitive compared to the ICET

3.1 Feasibility: battery capacity and depot charging power

BETs and AETs are assumed to have a lighter powertrain compared to the ICET which results in a lower vehicle body weight. The BET weight differential is constant between experiments at 2 t, while for the AET, it varies between 1.9 and 3.5 t depending on ADS weight and potential weight reductions from not having a driver’s cabin. The payload capacity is a result of the initial vehicle body weight differential, the allowed max GVW and the battery capacity and gravimetric energy density. The upper part of Fig. 5 shows the relation between battery capacity and the difference in payload capacity between the BET and ICET (left) and AET and ICET (right). The minimum battery capacity to break even in payload capacity is around 530 kWh for the BET and 545 kWh for the AET. If not accounting for the two-tonne increase in GVW for electric trucks, the payload break-even would instead be at the dashed horizontal line. Differences in the assumed initial vehicle body weight differential would similarly be manifested by shifting this line upwards (lower weight differential) or downwards (higher weight differential). For CST2 and CST3, a large share of experiments does not inflict any payload reduction, and almost none by more than 20%. Experiments where the truck grosses out are indicated by triangle markers in Fig. 5. These experiments are excluded from the subsequent analyses.

Fig. 5

Top: Relationship between battery capacity and difference in payload capacity compared to ICET for BET (left) and AET (right) across experiments. A negative value indicates a loss in payload capacity, and vice versa. The dashed line represents the break-even point if the additional weight allowance is not considered. Bottom: Distribution of payload capacity differential across experiments by charging strategy type

The required depot charging power for the BET is less than 300kW in 98% of the experiments that are not grossing out. The corresponding share for AET is 99%. However, around 20% and 13% of experiments for CST1 require depot charging power > 300kW for BET and AET respectively, see supplementary materials (part C) for details. When accounting for both feasibiltiy criteria, the number of infeasible experiments is 745 for BET and 470 for AET.

3.2 Cost performance, productivity, and utilization

The total costs per tkm, ${C}_{tk{m}_{t,e}}$, is partly determined by total number of tkm the truck operates over its lifetime, i.e. its productivity. Figure 6 illustrates this relationship for BETs (top) and AETs (bottom) by charging strategy type for a random sample of 100 scenarios. The black line indicates a best fit curve of the ICET cost performance results across all feasible experiments, for comparison. While there is a clear negative correlation between ${C}_{tkm}$ and total productivity, there is a large variation in ${C}_{tkm}$ for both BETs and AETs, in particular for CST2 and CST3. As an example, for CST3 there are many experiments with a low productivity but also a low ${C}_{tkm}$. These are likely cases when the battery can be heavily utilized, in combination with a high utilization of the charging infrastructure, along with favorable other scenario parameters, such as a high diesel price.

Fig. 6

Relationship between ${C}_{tk{m}_{t,e}}$ and total tkm for BET (top) and AET (bottom) per experiment for a random sample of 100 scenarios. Each marker represents one experiment. The black line is a trend line for the ICET results across experiments. The color coding denotes the value for ${u}_{c,othe{r}_{depot,destination}}$, used as a proxy for charging infrastructure utilization. The marker type shows whether the total battery utilization in EFCs is below or above the median value in each CST for BET. For AET, the marker type shows whether the remote operations cost, used as a proxy for operator utilization, is below or above the median

The productivity might differ between the BET/AET and the ICET within experiments. It is therefore relevant to analyze how such differences relate to the cost performance, ${\rho C}_{tkm}$. This relationship is illustrated in Fig. 7. A higher relative productivity for BET or AET does not clearly correlate with a better cost performance and there are many experiments where the productivity of the BET or AET is lower than the ICET productivity, but still being cost competitive (${\rho C}_{tkm}$<0). This suggests that it is mainly other factors that drive cost performance.

Fig. 7

Relationship between ${\rho C}_{tkm}$ and relative total tkm compared to the ICET for BET (top) and AET (bottom) for a random sample of 100 scenarios (same sample as in Fig. 6). The dashed black line represents the best-fit linear trend across experiments to support interpretation. The color coding denotes the value for ${u}_{c,othe{r}_{depot,destination}}$. The marker type shows whether the total battery utilization in EFCs is below or above the median value in each CST for BET, and for AET whether the remote cost is below or above the median

3.3 Impacts of charging strategy choice on cost performance

Figure 8 visualizes what types of scenarios are feasible to operate with different CSTs (top) and the share of scenarios that are both feasible and cost competitive (bottom) over different shuttle distance and payload parameter bin combinations. For brevity, only BET results are reported since the pattern is highly similar for AETs. CST1 is not feasible in many scenarios with high payload and long distance, but most scenarios that are feasible would also be cost competitive. For CST2 and CST3, most scenarios are feasible to operate.

Fig. 8

Share of scenarios where BET is feasible (top) and the share of feasible and cost competitive scenarios (bottom) for various shuttle distance and payload combinations, separated by charging strategy type (CST)

Which CST that yields the best cost performance differs between scenarios. Figure 9 shows the share of all feasible scenarios in which the respective CST yields the best BET cost performance. CST1 is generally the best strategy for shuttles of less than 325km. For longer distances, CST2HP often is the best option while CST3HP is the most competitive for the longest and heaviest shuttles. Note that these results also factor in whether the CSTs are feasible for each scenario.

Fig. 9

Share of scenarios within various shuttle distance and payload combinations for which the corresponding charging strategy type (CST) has the best BET cost performance out of the set of feasible CSTs

How important is it then to choose the charging strategy with the best cost performance? Fig. 10 shows the relationship between ${\rho C}_{tkm}$ for the best strategy (x-axis) and the delta in ${\rho C}_{tkm}$ between the best strategy and the average cost over all feasible strategies (y-axis) per scenario. This delta can be interpreted as the expected regret from randomly choosing between feasible strategies instead of the best feasible strategy. In 80% of scenarios, the average strategy cost performance is less than 10p.p. worse than for the best strategy. However, for both BET and AET, there are 13% of scenarios in which using a random feasible strategy instead of the best strategy would yield an uncompetitive solution instead of being competitive. Such cases are marked by red vertical lines in Fig. 10.

Fig. 10

Relationship of ${\rho C}_{tkm}$ for the best scenario (x-axis) and the increase in ${\rho C}_{tkm}$ averaged across feasible strategies compared to the best strategy (y-axis). Each marker represents a scenario and is color coded corresponding to what CST is the best option. The vertical line is red in cases where randomly choosing strategy instead of the best one would turn a cost competitive case into a non-competitive case. The dashed diagonal line represents the boundary for cost performance, which is in the area below and to the left of this line

3.4 Identification of robust profitability conditions

In addition to the previously analyzed aspects of utilization, productivity, and charging strategy, there are many other parameters that influence BET and AET cost performance between experiments. Figure 11, illustrates the result of a dimensional stacking analysis which is used to identify input parameter regions with a high share of cost competitive experiments (see supplementary materials part D for details). The color of each cell in Fig. 11 illustrates the share of cost competitive experiments for a specific bin combination of the four parameters with the highest influence on cost performance. For BET (left in Fig. 11), a combination of high charging infrastructure utilization by other trucks and high diesel price yields a high share of cost competitive experiments, regardless of other parameter values. Conversely, if the charging infrastructure utilization by other trucks is low, BETs are rarely cost competitive. For AETs, there is a similar pattern where most experiments with a high diesel price in combination with a high charging infrastructure utilization by other trucks are competitive. The cost for remote operations and ADS costs are also important parameters.

Fig. 11

Dimensional stacking to identify parameter regions with high concentrations of cost competitive experiments for BET (left) and AET (right)

4 Discussion

The results indicate that the type of shuttle flows studied in this paper are well-suited for electric truck operations since BETs and AETs would be feasible and cost competitive to operate in many scenarios. This is, however, conditionalized on that trucks with appropriate battery capacities can be sourced, and the technical and legal feasibility of AET operations. For longer flows and with heavier payloads, it might also be important that there is access to appropriate destination charging. The results are influenced by the scenario parameter ranges which account for potential future technological developments which favor the BET and AET performance compared to the current state. Nonetheless, this finding is in line with previous work which has highlighted repetitive shuttle flows as highly suitable for electric trucks [16].

This study provides insights in how differences in productivity over the truck life between BETs and AETs compared to ICETs relate to their cost performance. This is in contrast to most previous studies which tend to assume equal productivity between vehicle types [16, 23, 28]. However, there is no clear connection between high relative productivity and cost performance for BET or AET in the results of this study which may indicate that it is high asset utilization rather than high productivity per se which primarily affects cost performance. Future research would benefit from incorporating more sophisticated models of battery degradation and to further investigate utilization rates and residual value of AET components.

A merit of this study is the comparison between different types of charging strategies for the same transport application. Most scenarios can be cost competitively served by CST1 unless the payload is too heavy, and the distance is above 325 km (Fig. 8). With CST2 and CST3, almost all scenarios can be served without grossing out, but with somewhat less robust cost performance. For many scenarios, there are only minor differences between choosing the best charging strategy and randomly choosing any of the feasible charging strategies. Karlsson and Grauers [16] proposed that shuttle flows are best served by relying only on depot charging. This is largely supported by that CST1 tend to be a cost competitive option in most scenarios as long as the battery weight does not cause the truck to gross out (Fig. 8), and that it is the charging strategy type with the best cost performance in a majority of scenarios (Fig. 9). However, for flows longer than 325km, CST2 might be the best option. CST3 is less often the best charging strategy type which suggests that for shuttle flows, reducing the battery capacity, and thus cost, by using en route charging rarely compensates for the additional charging costs. By using dimensional stacking, the utilization rate of chargers from other trucks was identified as a key factor for determining cost performance. To better understand what charging utilization rates are realistic for various types of transport applications, there is a need for fleet level studies using realistic transport demand data simulating sharing of charging infrastructure.

Nykvist and Olsson [23] identified a negative relationship between BET cost per tkm and total vehicle weight while Mauler et al. [21] suggested that BETs are less competitive for heavy payloads partly due to the opportunity cost from a reduced cargo capacity and therefore missing revenues from unfeasible high-payload missions. In this present study, there is no clear correlation between payload and BET or AET cost performance, but the payload weight is a key determinant for what scenarios and strategies that are feasible to operate. This highlights a key modeling choice, namely whether to model operations for specific transport missions, as in this study, or as a truck serving a share of the trucking market, as in Mauler et al. [21]. This choice depends on the intended use of the truck: is it intended for a long-term deployment to a specific set of transport missions, or is it intended to be deployed for a variety of flows or operating primarily on the spot market.

The following are the key limitations:

• The repetitive shuttle application that is studied has special characteristics, for instance a highly predictable and stable energy consumption between trips. Therefore, the findings may not generalize to other applications.

• Key constraints for current real-world availability of BETs and AETs are not considered by assuming that battery capacity can be freely chosen, and that AETs are technically and legally feasible to operate for the considered applications.

• Several key mechanisms are modeled simplistically. For instance, battery degradation, energy consumption, residual value, and costs for charging infrastructure and remote operations.

• The representation and analysis of AETs is highly simplified. A key limitation is the use of identical vehicle schedules for BET and AET which neglects opportunities for leveraging AETs arising from the lack of driver-related planning constraints and costs [5, 6].

• The weight differential between electric and ICE powertrains is assumed constant between experiments. The 2 ton BET to ICE powertrain weight differential is higher than in some previous studies [15, 19, 20], but lower compared to others (e.g. [21, 27]). For illustration, adopting the powertrain weight parameters from other studies to the 400kW powertrain assumed for this study yields a weight differential from 1.2 tonne in the lower end [19], and up to 2.8 tonne in the higher end [27].

• There is no representation of actual transport demand, which limits the potential for modeling the benefits of increased productivity or comparing different set ups for serving the demand, for instance by using a smaller AET vehicle operating more frequently than the BET and ICET.

• Identification of parameter sub spaces is only done with dimensional stacking. There are several additional techniques that could provide complementary insights such as regional sensitivity analysis [25] and scenario discovery using the Patient Rule Induction Method [7]

5 Conclusions

This paper analyses BET and AET cost performance for medium and long-distance shuttle flows in Sweden using different types of charging strategies, and how it varies over uncertain market and technology conditions. An overarching finding is that shuttle flows are highly electrifiable. For most scenarios, there is at least one charging strategy that enables the shuttle to be served by a BET or AET without violating vehicle weight regulations. In many scenarios where the payload and distance are low to moderately high, it is possible to rely only on depot charging between shifts. A large share of the feasible scenarios would be cost competitive for a BET or AET to operate, in many cases regardless of what type of charging strategy is being used.

The cost performance of both BETs and AETs benefits from a high utilization of key assets such as the battery and charging infrastructure. For AETs, effective use of remote operators is also important. A high utilization of these components is, in addition to the price of fuel and energy efficiency of trucks, a key determinant of cost performance. All in all, this suggests that one key for cost effective utilization of shuttle flow is to carefully select the battery size and charging strategy based on the intended operations and effective sharing of charging infrastructure between multiple trucks.

In addition to the shuttle flow-specific findings, this study contributes by demonstrating how computational experiments, and techniques for high-dimensional data analysis can be used to leverage modeling of the economics of electrified heavy-duty road freight. It also identifies two areas for future research: developing and integrating more sophisticated models of the degradation and residual values of BETs, AETs, and batteries, and studies on fleet level effects of BET and AET operations from a transport demand perspective.