Modes#
Vehicle Disposal Model#
Researchers hypothesize that households may let go of their nth vehicle, and in urban areas with reliable public transit, dedicated pedestrian & bicycle infrastructure, and affordable & reliable SAV service, select households may no longer own a vehicle. A vehicle disposal model developed by a stated preference survey in Florida is used in our simulations, in addition to baseline vehicle ownership data (Menon et al., 2019). There are two random parameters ordered probit model for one-vehicle and multi-vehicle households (Table 1 & Table 2, respectively). The result is that vehicle count in the region due to households relinquishing vehicles in response to TNCs. Households have a random draw from a triangular distribution (A = 0.5, B = 1.0, Mean = 0.7) to determine the exact number of disposed vehicles (for multi-vehicle households). Modelers can alter this distribution to increase the propensity for multiple disposed vehicles. Modelers can also alter the threshold parameters corresponding to the Likert scale expressing a respondent’s likelihood of letting go of a vehicle (1=extremely unlikely and 5 = extremely likely) or the triangular distribution parameters. However, the baseline model should still capture some disposal, since it is subject to TNC service characteristics like fare and fleet size, as well as that of the household (size and income), the primary driver (gender, ethnicity, education status) and their typical driving needs (travel duration, parking). An example of a POLARIS simulation using this model is Gurumurthy et al. (2020).
The utility equations for the single- and multi-vehicle households is generalized in Equation 5. Total home accessibility is the sum of accessibility to employment by sector type (e.g., Government, Services, Other, Manufacturing, Retail, and Industrial).
U=Constant+βMale * Male+…… + βAccessibility * (Total Home Accesibility) + βFleetSize * ((Fleet Size Scenario-Fleet Size Base)/(Fleet Size Base))
βTNCCost * ((Base Fare Scenario-Per Mile Cost ScenarioVMT/365)-(Base Fare Base+Per MIle Cost BaseVMT/365))
Table 1: Single-vehicle household random parameter ordered probit model (adapted from Menon et al., 2018)
Variable description |
Coefficient (St. Dev) |
t-stat |
|---|---|---|
Constant |
0.1 |
6.50 |
Male |
-0.211 (1.627) |
-1.61 (12.38) |
Millennial |
0.679 |
4.54 |
Post Graduate Degree |
0.119 (0.821) |
0.92 (7.43) |
Household Size Greater Than 3 |
0.935 |
4.21 |
Singe Licensed Driver Household |
-0.258 (1.456) |
-1.83 (12.06) |
Moderate Commute Distance (<10 mi) |
-0.231 (1.221) |
1.70 (9.98) |
High Daily Travel Time (>90 min/daily) |
-0.662 (2.150) |
-3.44 (9.64) |
Low Parking Time (<5 min) |
-0.592 |
-4.36 |
Crash History |
0.101 (1.239) |
0.70 (12.60) |
Crash History of Totaled Vehicle |
-0.424 |
-2.52 (7.32) |
Accessibility |
0.08 |
|
Fleet Size |
1.0 |
|
TNC Cost |
-0.01 |
|
Threshold 1 |
2.168 |
13.55 |
Threshold 2 |
3.406 |
16.93 |
Threshold 3 |
5.308 |
17.36 |
Number of observations |
417 |
|
Log-likelihood at convergence |
-581.017 |
|
Restricted log-likelihood |
-607.209 |
Table 2: Multi-vehicle household random parameter ordered probit model (adapted from Menon et al., 2018)
Variable description |
Coefficient (St. Dev) |
t-stat |
|---|---|---|
Constant |
0.08 |
6.45 |
Male |
0.119 (0.622) |
1.49 (11.41) |
Millennial |
0.593 |
4.33 |
White |
-0.346 |
-3.03 |
Post Graduate Degree |
0.305 |
3.76 |
Singe Licensed Driver Household |
-0.706 |
-4.47 |
Vehicle Ownership Greater Than 3 |
-0.289 |
-3.54 |
Moderate Commute Distance (<10 mi) |
0.362 (0.386) |
3.70 (4.50) |
High Daily Travel Time (>90 min/daily) |
0.174 (0.926) |
1.54 (8.26) |
Low Parking Time (<5 min) |
-0.184 |
-2.18 |
Crash History |
0.272 (0.538) |
2.33 (11.26) |
Crash History of Totaled Vehicle |
-0.165 (0.646) |
-1.52 (7.45) |
No Injury in Crash |
-0.210 |
-2.14 |
Accessibility |
0.01 |
|
Fleet Size |
0.8 |
|
TNC Cost |
-0.008 |
|
Threshold 1 |
0.816 |
15.14 |
Threshold 2 |
1.548 |
22.55 |
Threshold 3 |
2.737 |
28.03 |
Number of observations |
797 |
|
Log-likelihood at convergence |
-1195.938 |
|
Restricted log-likelihood |
-1238.243 |
SOV versus HOV assumptions#
Within the mode choice methods file, there is a function called “defined travel context.” We set the availability of modes to be true or false depending on household properties. For example, if there are no vehicles in the household, we automatically assume transit is available and that SOV or HOV modes are unavailable. For some zero-vehicle households we assume that they can call on friends for a vehicle 25% of the time, but there is no subsequent check whether the owner or borrower is using that vehicle for a trip when the auto mode is chosen for a new trip. If the person is under 16 (a child), then the mode chosen is in a unique function called “choose mode for child.” Here it takes into account the child’s age, whether they live in a city, and the distance between home and school to determine probability of choosing HOV, walk, bike, bus, or school bus. In contrast, the activity methods file has a function called “add activity to schedule event handler” that checks for escort trips when a child is less than 6 years old. For example, if there is no adult but there is to/from a school activity then the child is forced to ride a school bus. If there is no adult but the child is set to return home from a non-school location, then they take a ghost HOV trip. For all unlicensed drivers the mode is always HOV otherwise.
There are also constraints for autos (and bikes) for new trip that are not home-based work trips. For example, if the person is not at their home or work/school location and the mode taken was neither a vehicle or a bike we assume that they do not magically have access to these modes. Although bike-sharing or car-sharing (or pick-up with HOV modes) are all possible in the real-world. Likewise, if the previous location was at work/school and they did not travel to this location with an auto then they do not have access to an auto. There is a tour-based constraint as well for when both the location and mode is planned and the next location is not at home. Here, the next tour is taken with the same mode.
The default option within POLARIS is HOV, unless there are no vehicles in the household at which the default is taxi. The mode choice model overrides the default of course, but there are no checks for whether a chosen individual’s choice of HOV is practical. POLARIS does allow HOV trips that are non-child trips but it is not HOV in the sense that like people that otherwise would have taken SOV are merged into an HOV trip (i.e, 1 SOV for person 1 and 1 HOV trip for person 2 becomes 1 HOV trip for persons 1 & 2). Anyone can choose an HOV mode from the utility mode choice model (even if there are no other passengers/drivers) and a household vehicle is free. Essentially HOV is like any other mode in that regard. If there is one household vehicle and the vehicle is already taken, another household member may start a trip and choose SOV. Since there is no vehicle available the person is forced into HOV (and is teleported essentially).
Non-motorized mode assumptions#
The section above has rules for mode availability and applies to bikes as well. For example, if the household does not have bikes, then bikes are not available. If an auto-based mode is less than 250 meters, then we assume that the person walks.
Transit#
POLARIS supports very detailed public transportation (or transit) modeling, owing to the creation of POLARIS around the City of Chicago – with rail, bus, park-and-ride infrastructure, among other transit mode and infrastructure types. The supply file reads in transit infrastructure such as rail-based links and stations, bus lines and stops, and GTFS scheduling information (see “Transit and active transport” section for more information).
*insert information here about how routes & schedules & stations are simplified.
*insert information about park-and-ride, kiss-and-ride, etc.
*insert link up to Yantao’s FMLM section from here too.
EVs#
In addition to decision-to-charge parameters and flexible charging policies that are mentioned in the Charging SAEVs section, EV charging and discharging (through consumption) is also an integral part of this powertrain type. EVs charge nonlinearly and charging efficiency is not constant during the charging process, especially at the extremes of the battery level. Modeling this complex charging behavior is difficult (even under constant voltage constant current charging). A simplification in large-scale analyses is a linear charging rate constrained by minimum and maximum SoC. The thresholds that limit the designed capacity of a battery help to prevent enhanced degradation of battery performance because of higher charging stress in boundaries of SoC (Argue, 2019). Thus, modelers often assume a protection buffer of 10-20%, especially for Level 3 charging. While charging and discharging rates are governed by C-rates (Collin et al., 2019), large-scale transportation models have assumed either a constant consumption rate for an average EV or estimated a simple split linear equation based on estimated charging duration (e.g., Loeb et al., (2018)). In POLARIS, charging was initially assumed to follow a simple linear rate (in units of miles per minute of charging) based on the charger speed but now is based on a linear regression equation. As electricity consumption (or discharge) is a function of the vehicle’s auxiliary power (e.g., on-board computers, HVAC, etc.) and the vehicle’s trajectory on different transportation facilities (e.g., high-speed control-access freeways versus low-speed arterials and collectors with signals), a sufficiently accurate electricity consumption model strengthens the validity of results from SAEV operations. Demir et al. (2014) categorize energy discharge models into three approaches: factor, macroscopic, and microscopic models, as reported for the electric vehicle routing problem (EVRP) in Basso et al. (2019). The factor model is the most simplistic and assumes a uniform energy discharge in kWh per unit distance traveled. Thus, the total energy consumption for a trip is the sum of energy discharged from the battery along each link on the route (see Iacobucci et al., (2018b, 2018b), Bauer et al. (2018)). Vosooghi et al. (2020) implemented an energy consumption model to calculate battery discharge, which does not appear to have an effect on fleet performance – percent eVMT in the range of 18.3% to 22.8% matches other studies but an average wait time between 13.2 and 13.9 minutes is high – however, this is likely a result of no absolute maximum wait time or relocation strategy. POLARIS, like other agent-based models (Chen et al., 2016; Loeb et al., 2018; Loeb and Kockelman, 2019), expresses battery capacity in mile-equivalents and not kWh in the framework. This macroscopic approach yields a more accurate consumption model although increases computational complexity. Microscopic models, like Fiori et al. (2016), that are leveraged in SAEV work by Sheppard et al. (2017) approximate discharge under different driving cycles. Although current EVs exhibit greater efficiency on city routes than freeways (Wu et al., 2015), eco-routing and smoother vehicle handling benefits arising from automation may minimize large differences in efficiency, which could reduce the need for microscopic consumption models. Additionally, many EVRP/ green VRP studies, use a simple energy consumption model linear with distance (Macrina et al., 2019). A researcher at ANL has implemented a machine learning model for EV consumption, which can be used instead of the factor-based approach. Please see Moawad et al. (2022). The outputs from the 24-hour simulation can be used as inputs into Autonomie to capture different emissions from the transportation network. The inputs include routed travel distance and vehicle type (using EPA powertrain types).
##EV Citations
Important
Please reference the following paper for more details on the machine learning EV consumption model(available here):
Moawad, A., Gurumurthy, K.M., Verbas, O., Li, Z., Islam, E., Freyermuth, V., and Rousseau, A.
A Deep Learning Approach for Macroscopic Energy Consumption Prediction with Microscopic Quality for Electric Vehicles.
Under review for publication in Transportation Research Part D.
Other references in this text:
Argue, C., 2019. What can 6,000 electric vehicles tell us about EV battery health? [WWW Document]. Geotab. URL https://www.geotab.com/blog/ev-battery-health/ (accessed 5.4.20).
Basso, R., Kulcsár, B., Egardt, B., Lindroth, P., Sanchez-Diaz, I., 2019. Energy consumption estimation integrated into the Electric Vehicle Routing Problem. Transp. Res. Part Transp. Environ. 69, 141–167. https://doi.org/10.1016/j.trd.2019.01.006
Bauer, G.S., Greenblatt, J.B., Gerke, B.F., 2018. Cost, Energy, and Environmental Impact of Automated Electric Taxi Fleets in Manhattan. Environ. Sci. Technol. 52, 4920–4928. https://doi.org/10.1021/acs.est.7b04732
Chen, T.D., Kockelman, K.M., Hanna, J.P., 2016. Operations of a shared, autonomous, electric vehicle fleet: Implications of vehicle & charging infrastructure decisions. Transp. Res. Part Policy Pract. 94, 243–254. https://doi.org/10.1016/j.tra.2016.08.020
Collin, R., Miao, Y., Yokochi, A., Enjeti, P., von Jouanne, A., 2019. Advanced Electric Vehicle Fast-Charging Technologies. Energies 12, 1839. https://doi.org/10.3390/en12101839
Demir, E., Bektaş, T., Laporte, G., 2014. A review of recent research on green road freight transportation. Eur. J. Oper. Res. 237, 775–793. https://doi.org/10.1016/j.ejor.2013.12.033
Fiori, C., Ahn, K., Rakha, H.A., 2016. Power-based electric vehicle energy consumption model: Model development and validation. Appl. Energy 168, 257–268. https://doi.org/10.1016/j.apenergy.2016.01.097
Iacobucci, R., McLellan, B., Tezuka, T., 2018b. The Synergies of Shared Autonomous Electric Vehicles with Renewable Energy in a Virtual Power Plant and Microgrid. Energies 11, 2016. https://doi.org/10.3390/en11082016
Loeb, B., Kockelman, K.M., 2019. Fleet performance and cost evaluation of a shared autonomous electric vehicle (SAEV) fleet: A case study for Austin, Texas. Transp. Res. Part Policy Pract. 121, 374–385. https://doi.org/10.1016/j.tra.2019.01.025
Loeb, B., Kockelman, K.M., Liu, J., 2018. Shared autonomous electric vehicle (SAEV) operations across the Austin, Texas network with charging infrastructure decisions. Transp. Res. Part C Emerg. Technol. 89, 222–233. https://doi.org/10.1016/j.trc.2018.01.019
Macrina, G., Di Puglia Pugliese, L., Guerriero, F., Laporte, G., 2019. The green mixed fleet vehicle routing problem with partial battery recharging and time windows. Comput. Oper. Res. 101, 183–199. https://doi.org/10.1016/j.cor.2018.07.012
Sheppard, C., Waraich, R., Campbell, A., Pozdnukov, A., Gopal, A.R., 2017. Modeling plug-in electric vehicle charging demand with BEAM: the framework for behavior energy autonomy mobility. Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States). https://doi.org/10.2172/1398472
Vosooghi, R., Puchinger, J., Bischoff, J., Jankovic, M., Vouillon, A., 2020. Shared autonomous electric vehicle service performance: Assessing the impact of charging infrastructure. Transp. Res. Part Transp. Environ. 81, 102283. https://doi.org/10.1016/j.trd.2020.102283
Wu, X., Freese, D., Cabrera, A., Kitch, W.A., 2015. Electric vehicles’ energy consumption measurement and estimation. Transp. Res. Part Transp. Environ. 34, 52–67. https://doi.org/10.1016/j.trd.2014.10.007
You should also refer to the citations page for a full listing of other important papers.