Mobility Energy Productivity Evaluation of Prediction-Based Vehicle Powertrain Control Combined with Optimal Traffic Management

Full text

2022-01-0141 Published 29 Mar 2022
Mobility Energy Productivity Evaluation of
Prediction-Based Vehicle Powertrain Control
Combined with Optimal Trac Management
Farhang Motallebiaraghi Western Michigan University
Kaisen Yao and Aaron Rabinowitz Colorado State University
Christopher Hoehne, Venu Garikapati, Jacob Holden, and Eric Wood National Renewable Energy Laboratory
Suren Chen Colorado State University
Zachary Asher Western Michigan University
Thomas Bradley Colorado State University
Citation: Motallebiaraghi, F., Yao, K., Rabinowitz, A., Hoehne, C. et al., “Mobility Energy Productivity Evaluation of Prediction-
BasedVehicle Powertrain Control Combined with Optimal Traffic Management,” SAE Technical Paper 2022-01-0141, 2022,
doi:10.4271/2022-01-0141.
Abstract
Transportation vehicle and network system eciency can
bedened in two ways: 1) reduction of travel times
across all the vehicles in t he system, and 2) reduction in
total energy consumed by all the vehicles in the system. e
mechanisms to realize these eciencies are treated as indepen-
dent (i.e., vehicle and network domains) and, when combined,
they have not been adequately studied to date. is research
aims to integrate previously developed and published resea rch
on Predictive Optimal Energy Management Strategies
(POEMS) and Intelligent Trac Systems (ITS), to address the
need for quantif ying improvement in system eciency resu lting
from simultaneous vehicle and network optimization. POEMS
and ITS are partially independent methods which do not
require each other to function but whose individual eective-
ness may beaected by the presence of the other. In order to
evaluate the system level eciency improvements, the Mobility
Energy Productivity (MEP) metric is used. MEP specically
measures the connectedness of a system while accounting for
time and energy externalities of modes that provide mobility
in a given location. A SUMO model is developed to reect real
trac patterns in Fort Collins, Colorado and data is collected
by a probe SUMO vehicle which is validated against data
collected on a real vehicle driving the same routes through the
city. Individual vehicle and system level eciencies are calcu-
lated using SUMO outputs for scenarios which integrate
POEMS and ITS independently as well as jointly. Results from
application of POEMS and ITS show improvement in energy
consumption and travel times respectively when compared to
the respective baseline scenarios. Our conclusion is that there
are promising synergistic benets to travel time and energy
eciency when POEMS and ITS are combined.
Introduction
The fuel-based transportation system has a signicant
impact on human interactions with the environment
and our nation's economy. When compared to other
modes of transportation such as aircra, rail, and marine,
road-based travel is responsible for the greatest share of CO2
emissions, greenhouse gas (GHG) emissions, and energy
usage. Vehicles transport 11 billion tons of freight and passen-
gers over 3 trillion vehicle-miles annually. e transportation
sector accounts for over 30% of total US energ y consumption,
and the average US household spends more than 15% of total
family expenditures on transportation, making it the costliest
expense category aer housing [1].
Knowing that transportation emissions surged more than
emissions from any other sector over the last thirty years,
transportation emissions must bea primary focus of mitiga-
tion eorts [2, 3]. Employing emerging techniques to mini mize
the environmenta l consequences of road-based transportation
can thus go a long way into mitigating the entire environ-
mental i mplications of the tra nsportation industr y. Connected
and automated vehicle (CAV) technology and electrication
are two examples with signicant promise to reduce emissions
Received: 31 Jan 2022 Revised: 31 Jan 2022 Accepted: 27 Jan 2022
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL 2
from road-based transport and facilitate decarbonization of
the transportation sector. Aside from improvements in
powertrain technology, recent automotive sector develop-
ments have resulted in considerable advancements in CAV
technology and improved vehicle control strategies. Vehicle
connectivity and automation are distinct technologies that
can exist separately from one another yet have signicant
synergistic characteristics. e ability of a vehicle to commu-
nicate information with other cars and infrastructure is
referred to as connectivity. Vehicle-to-vehicle (V2V), vehicle-
to-infrastructure (V2I), and other cooperative communica-
tions networks can provide this functionality [4-11]. Vehicle
automation refers to any scenario in which management of a
vehicle capacity that would typically bemonitored by a human
driver is transferred to a computer which is referred to as
Advanced Driver Assistance System (ADAS). Examples of
ADAS are Cruise control, adaptive cruise control, active lane-
keep assist, and automated emergency braking that may
befound in today's vehicles.
Vehicle electrification is another key component to
improve fuel economy and reduce emissions. Hybrid Electric
Vehicles (HEV) and Plug-in Hybrid Electric Vehicles (PHEV)
have recently received a lot of attention due to their higher
potential to improve fuel economy (FE) and emissions
compared to traditional Internal Combustion Engine (ICE)
vehicles [12]. However, there is still scope for enhancing the
performance of the current generation of HEVs [13]. Recent
application of CAV technology shows improvement in energy
eciency on HEVs and PHEVs through implementation of
Predictive Optimal Energy Management Strategies (POEMSs)
[14-19].
POEMS is an application of optimal control where a
mathematical optimization problem is formulated by dening
the mass of fuel used as a cost to beminimized over a xed
drive cycle [14]. POEMS can beformulated as a real time
control using either Pontryagin's Minimum Principle (such
as in ECMS [20]) or using Dynamic Programming (DP) over
a limited prediction time horizon. e application of DP over
a limited prediction time horizon is referred to as Model
Predictive Control (MPC). In either case predicted future
vehicle velocity is used as an input to the strategy [6, 21].
POEMS performance is, thus, dependent on the quality of
future vehicle velocity predictions. In recent years, vehicle
connectivity and sensing technologies as well as advances in
Machine Learning (ML) and Artificial Neural Network
(ANN) technology have made high delity predictions feasible
[22-25] enabling the implementation of POEMS.
Intelligent technolog y has become a key solution to trac
control in urban cities to mitigate congestion and reduce
vehicle delay. Intelligent trac signal control eorts primarily
focus on applying dierent algorithms to meet the needs of
trac safety and eciency with the data from monitoring
cameras, road sensors, existence detectors, etc. [26]. Typical
modes of operation for trac signals include pre-timed (xed
time) and actuated operations. Despite being straightforward
and easy to coordinate between intersections, pretimed opera-
tion is more cost-eective for close intersections with constant
traffic volumes but lacks flexibility to adjust with traffic
demand and possibly causes excessive vehicle delays. Actuated
operation, on the other hand, can adjust phase durations and
sequences by detecting real-time trac conditions, including
prolonging or shortening phase durations and skipping a
phase based on trac demand. However, actuated operation
does not have a xed cycle length and is therefore hard to
coordinate among intersections.
Trac signal designs are critical to trac safety and e-
ciency at intersections during day-to-day service including
normal or heavy trac scenarios [27]. Signicant research
eorts have been put forth to study signal timing optimization
techniques to mitigate recurring trac congestions during
rush hours. An improved automatic trac volume detection
technique using V2I communication was proposed to get the
trac information in time for the following optimizations
[28]. Discrete dynamic optimization models for optimal cycle
length and green time allocation were evaluated to identify
the most appropriate design to deal with congested trac
scenarios [29]. In recent years, there have been some emerging
research eorts investigating intersection signal designs for
non-recurrent congestion. A cell transmission model for a
signalized intersection was developed for dierent congestion
evacuation schemes [30]. GPS data was utilized for a global
network modeling to evaluate the trac condition with matrix
factorization and clustering methods during emergency
recovery [31]. A signal timing optimization model using queue
length as the penalty value has also been developed under
trac incident scenarios, in which a heuristic algorithm
(simulated annealing algorithm) was adopted [32].
With the need to minimize energy and time expended
during travel also comes the need to quantify the benets of
doing so in a united and holistic way. e Mobility Energy
Productivity (MEP) metric developed by researchers at the
National Renewable Energy Laboratory (NREL) measures
travel accessibility to opportunities with weightings for travel
time, travel cost, and energy use at any given location [33, 34].
MEP has been constructed as a theoretically grounded but
simplistically presented metric to help cities and transporta-
tion planners understand holistic impacts of novel and
emerging transportation strategies and technologies on the
quality of mobility in their jurisdictions. A high MEP score
equates to more productive and energy ecient mobility, or
more simply, a greater access to opportunities in the context
of cost, time, and energy of modes that provide mobility in a
location. MEP is used as an evaluation metric for this study
as it is desirable to exemplify potential synergistic travel time
and energy benets that this study explores.
Many existing studies use a combination of energy
management strategies (EMS) with Intelligent Trac Systems
(ITS). Most of these studies used eco-driving or POEMS with
real world trac data [35-39]. Only a few studies investigated
the eect due to synergic application of EMS and optimized
Trac Management Systems (TMS) on fuel economy and
trac eciency [40, 41]. ere is a signicant research gap in
addressing the eect of application of POEMS and TMS
methods simultaneously on fuel economy and mobility. To
address this research gap, this paper investigates the eect of
POEMS and tra c signal timing optimization on each other’s
performance. e main contributions include:
•Simultaneous implementation of POEMS and adaptive
trac signal strategy: A validated SUMO model is used
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

3
MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL
along with trac signal and vehicle optimal controls
acting individually and simultaneously.
•Case studies development: Four case studies are
developed to evaluate the eect of aforementioned
energy and trac optimization on each other’s
performance at the corridor in Fort Collins, CO.
•Performance assessment using the MEP metric: MEP
metric which documents the combined benets of
improved energy eciency from POEMS with reduced
travel time and improved energy eciency from optimal
trac signal timing is utilized as an assessment metric.
Methodology
In this section, the experimental method of drive cycle devel-
opment and real-world data collection is explained. en
evaluated FE and TMS optimization methodologies as well
as MEP metric/data overview are discussed. At the end an
overview of all studied and developed case studies
are presented.
Drive Cycle Development
Over the course of two days, one driver gathered a total of 13
drive cycles. e chosen driving cycle was a 4-mile drive cycle
through urban arterial highways in downtown Fort Collins,
Colorado. Data gathering followed driving from the College
Mulberry intersection to Shield Mulberry intersection to
Shield Prospect intersection to College Prospect intersection
and back to origin point which is shown in Figure 1.
is drive cycle was selected with the intention that it
would begenerally representative of urban arterial driving.
In order to accomplish this, the drive cycle was selected and
tested in order to verify that it was suciently similar to the
Environmenta l Protection Agency (EPA) Urban Dyn amometer
Driving Schedule (UDDS) drive cycle. Figure 2 shows velocity
vs. time trace of one of the 13 collected laps of the data
drive cycle.
e similarity assessment of characteristics of the data
drive cycle and UDDS EPA dynamometer drive cycle was
published in a previous study [19].
FE Optimization
Predictive Optimal Energy Management Strategies (POEMS)
for HEVs employs forecasts of future vehicle velocity to deter-
mine an ideal powertrain control approach, resulting in
improved energy economy. A POEMS, as seen in Figure 3, is
made up of three primary subsystems.
1. e perception system, which anticipates vehicle
motion based on historical and present vehicle
motion, powertrain conditions, driver inputs, ADAS,
and V2X (Vehicle to Everything) data.
2. e planning subsystem, which calculates optimal
controls based on anticipated vehicle velocity.
3. e vehicle plant, which may beeither the actual car
or a high-delity simulation model of the vehicle.
e outputs of the system are the fuel consumption and
change in State of Charge (SOC) of the HEV battery.
Perception Subsystem In order for POEMS to beimple-
mentable in real world conditions, the future vehicle velocity
must bederived from actual predictions based on available
data rather than from prior knowledge of a vehicle velocity
trace. Data used in this prediction must also bedata which is
generally ava ilable to connected vehicles. A pract ica l taxonomy
of this data is given in Table 1.
FIGURE 1 Driving map for selected drive cycle.
FIGURE 2 Velocity vs. time trace of one drive cycle.
FIGURE 3 POEMS logic system [19].
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL 4
TAB LE 3 EPA dynamometer drive cycle FE (km/L)
comparison results from Toyota Prius model and Argonne
National Lab Downloadable Dynamometer Database [19].
Drive-Cycle Data Model
Percentage
Dierence
UDDS 32.14 31.79 1.09%
US06 29.72 30.30 1.95%
HWFET 19.26 18.98 1.45%
The use of various ML and ANN techniques in the
creation of future velocity predictions were considered and a
deep Long Short-Term Memory (LSTM) ANN was found to
bethe most eective. e reader is directed to prior publica-
tions for more discussion and information [4, 19, 42].
Planning Subsystem e planning system for HEV
POEMS can bestated as a classical optimal control problem
with the battery SOC and powertrain gear as the states, the
torque split and gear shi as controls, and the vehicle velocity
as an exogenous input. e objective is to minimize fuel
consumption while observing battery SOC constraints.
Solutions to the HEV fuel consumption minimization
problem follow two forms. e rst is a nominally real time
control scheme based on Pontryagin’s Minimum Principle
(PMP) such as ECMS and a-ECMS [43, 44] and derivative
methods. Although nominally real time control methods,
these still require future knowledge to compute an equivalence
factor which is analogous to a prediction [45]. e second form
is DP based methods which compute a globally optimal
solution based on a future velocity prediction. DP methods
explicitly require high-delit y predictions to funct ion properly
but the exact relationship between prediction quality and DP
eectiveness is not known [16].
In order for DP POEMS to beimplemented in real time
the method must beadjusted to use a limited horizon predic-
tion. e limited horizon prediction is used to generate the
optimal controls for the given time-step and the process is
repeated at each time step. is method is referred to as Model
Predictive Control (MPC). e reader is directed to the prior
publication [16] for more discussion and information.
Plant Subsystem In order to maximize the realism of the
FE results of the study, the plant subsystem used was a vali-
dated Autonomie model of a 2010 Toyota Prius. is model
is based on the Autonomie power-split HEV model with the
vehicle parameters shown in Table 2 derived from publicly
available information about the 2010 Toyota Prius. In lieu of
a manufacturer provided model, this Autonomie implementa-
tion of a Prius model is as accurate as can beattained.
With these va lues the model was run for the UDDS, US06,
and HWFET EPA dynamometer drive-cycles and the results
were compared against those found in Argonne National Lab ’s
Downloadable Dynamometer Database (ANL D3) [46]. e
results of this comparison are shown in Table 3.
Considering all the above listed discrepancies to bewithi n
the acceptable range, the Autonomie Toyota Prius model was
used for FE evaluation for this study.
Trac Optimization
Real-life drive cycles and simulated drive cycles are compared
to decide which car following models in SUMO (Simulation
of Urban MObility) are most realistic. Parameters for SUMO
vehicle conguration inputs are calculated based on real drive
cycles used for EMS. e parameters include acceleration rate,
maximum speed, deceleration rate, and emergency
deceleration rate.
DynamicPhaseSelection
andQueueLength
DissipationSequence
Algorithm
SUMO trac network according to real-life drive cycles is
built to demonstrate TMS optimization results and extract
drive cycles for EMS optimization. TMS optimization is
Dynamic Phase Selection combined with Queue Length
Dissipation algorithm (DPS+QLD), which dynamically
changes phase sequences and signal timing based on instant
trac volume change.
TAB LE 1 Data sources and associated signals from [19].
Data Source Signal Description
VEH General Vehicle
Signals
Signals such as speed,
acceleration, throttle position,
and steered angle which can
befound via CAN on any
vehicle
VEH Historical
Speeds (HS)
Historical speed data for the
vehicle at the current location
ADAS Lead Vehicle
Track (LV)
Relative location of confirmed
lead vehicle from ADAS system
V2I Signal Phase
and Timing
(SPaT)
Signal phase and timing of next
trac signal
V2I Segment Speed
(SS)
Trac speed through current
road segment
TAB LE 2 HEV vehicle parameters [19].
Parameter Value
Overall Vehicle Mass 1530.87kg
Frontal Area 2.6005 m2
Coecient of Drag 0.259
Coecient of Rolling
Resistance
0.008
Wheel Radius 0.317m
Final Drive Ratio 3.267
Sun Gear Number of Teeth 30
Ring Gear Number of Teeth 78
Battery Open-Circuit Voltage 219.7V
Battery Internal Resistance 0.373 Ω
Battery Charge Capacity 6.5 Ah
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

5
MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL
Dynamic Phase Selection
(DPS) Algorithm: Phase
Sequence Selection
Aer a trac congestion has occurred, typically very few or
no vehicles from the approach with minor roads use the green
light. erefore, such a period can beallocated to other phases
to improve the mobility of the remaining approaches to avoid
long queues. DPS can adaptively choose the best phase
sequence of a cycle to make the trac more ecient with the
help from monitoring detectors. Starting at the major road
movement, the next phase is chosen dynamically based on all
candidate phase options with the following algorithm [47]:
1. Compute the priority for each phase given in the list
of indices (the sequence of potential phases that will
beused for the next phase when the current one is
over) for next possible movements as 'next' attribute.
Priority is made according to the number of active
detectors for that phase. A detector is deemed “active”
when either of the following conditions is met:
a. e time gap between consecutive vehicles is shorter
than the threshold.
b. Vehicle existence is detected aer the signal has
turned to red from the last cycle.
2. e current phase is available to continue implicitly if
its maximum duration (MaxDur) is not reached. e
current phase detector gets a bonus priority.
3. e phase with the highest priority is used for the
next cycle over other possible movements.
4. If no trac is detected, the phases will follow the
default cycle dened by the rst value in the
'next' attribute.
5. If a particular phase needs to remain active for a no-
trac scenario, it must have a high maximum
duration value and its index number is on the
'nex t' list.
6. If the time that an active detector was not served
exceeds the preset time threshold, such a detector will
receive bonus priority of the time that was not served.
is could prevent those phases serving more trac
from being consistently served.
Based on the algorithm introduced above, DPS can
choose the next phase according to the real-time trac situ-
ation. In such a case, a certain amount of time could besaved
for better movement of the intersection for other phases.
Queue Length Dissipation
(QLD) Algorithm: Optimal
Green Light Calculations
Aer phase sequences are selected, queued-up vehicles on the
approach need longer green light time for congestion mitiga-
tion. However, the required green time should becalculated
based on the queue lengths of not only the current approach,
but also other approaches of the intersection at the same time
to avoid causing additional congestion in other directions.
erefore, a maximum green time gmax should beconsidered
to balance the green time allocations among different
approaches. Based on the analytical method by Akçelik [48,
49] , the average green time and cycle length of an actuated
controller adopt a xed unit extension setting by assuming
the arrival headway follows the bunched exponential distribu-
tion [50]. Existing vehicles remaining in front of the green
light are dened as bunched vehicles while new arriving
vehicles are dened as free vehicles. Dierent proportions of
bunched and free vehicles dene minimum and maximum
green time, gmin and gmax respectively. e green extension
time eg is set based on the queue length at the red light ending
time point, and the phase change does not happen during the
saturated portion of the green period.
e green time g can beestimated by the following equa-
tions introduced by Akçelik [49] :
gg e
sg

(1)
where g is the green time and gs is the saturated portion
of the green period, and eg is the extension time if the phase
change happens aer the queue clearance period.
e green time boundary is set as:
ggg
minmax
<<
(2)
e green light distribution for the approach with the
incident follows the rules considering the queue lengths of
other approaches [49]:
g
geforg g
gfor gg
sg
ss
j
sssj






,
, (3)
where
gsj = the saturation portion of the green period of other
directions and j = 1,2,3.
Trac signal optimization will beconducted in two
phases. During the rst phase immediately following the
incident, dynamic phase selection (DPS) is used for skipping
the unused phase of the blocked approach due to incidents to
save the time loss of the intersection operation. e second
phase is when the incident is cleared, and the queue length
information is collected to calculate the optimal signal timing
to dissipate the queue as soon as possible. When gmaxis reached,
the controller will move to the next phase to avoid redundant
green time causing long queue lengths on other approaches.
Aer the rst cycle, queue length information at the red end
time point is collected again for the following signal timing
calculations. e trac signal control optimization covers
both crash and recovery stages. DPS is used to skip unneces-
sary phases during crash stage, and the queue length dissipa-
tion algorithm is used to dissipate the queue length at crash
lanes as soon as possible. e whole process for this optimiza-
tion is called DPS+QLD. Assessment metrics for trac opti-
mization is segment average speed/travel time and is provided
as an input to calculate the MEP metric.
MEP Metric & Data Overview
To evaluate the impacts of these case studies on productive,
energy-ecient mobility, the MEP metric is calculated for
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL 6
each case using local land use data, travel data, and vehicle
characteristics. e MEP metric quanties access to six types
of opportunities (jobs, education, recreation, shopping,
healthca re, and mea ls). is can bedone for any given location
by computing travel time isochrones, proportioning oppor-
tunities access by engagement frequency, and weighting the
resulting opportunity measure by time, cost, and energy of
any mode that can beused to access the opportunities. is
calculation happens in three steps for each location the MEP
metric is being quantied at:
1. Quantify the cumulative number of opportunities
that a travel mode (e.g., car) can reach within a
certain travel time threshold.
2. Normalize all reachable opportunities by a) their
relative magnitude of occurrence (e.g., in a city, there
typically are a lower number of healthcare facilities
compared to retail stores); and b) their engagement
frequencies (e.g., healthcare opportunities are
engaged less frequently compared to grocery or
shopping opportunities). Engagement frequencies are
derived from the National Household Travel Survey
in 2017 (NHTS) [51].
3. Weight the cumulative opportunities measure using
time, energy, and cost decay functions such that more
weight is assigned to modes that provide faster,
aordable, and energy-ecient access
to opportunities.
e equation to compute MEP can be mathematically
described as follows:
ME
Po
oe
i
kt
iktik t
Mikt





10 · (4)
where
oikt = the opportunity measure, which represents the
number of opportunities that can bereached by
mode k in time t from location i; and Mikt is further
dened as:
Metc
iktk k


(5)
where:
Mikt = Composite decay function for ti me, energy, and cost
ek = the energy intensity in kWh per passenger-mile of
mode k
t = is the travel time in minutes
ck = the cost in dollars per passenger-mile of using
mode k
α, β, and σ= the weighting parameters
e weighting parameters for α(energy), β (time), a nd
σ(cost) are −0.5, −0.08, a nd −0.5, respectively. For more details
on the MEP methodology see [33, 34].
MEP can beapplied to multiple modes (e.g., walking,
transit, biking, driving), but for this analysis wefocus on
impacts to light-duty vehicles driving on the specied route
in Fort Collins, CO. e drive mode MEP is calculated using
a routable road network in Fort Collins with average vehicle
speeds for all road links in the region, and activities by type
are evaluated using parcel level third-party land use data.
Wetranslate fuel economy improvements and travel speed
improvements from simulations outlined in above sections
to the baseline real-world observed speeds by link and average
driving fuel economies to evaluate optimization scenarios on
MEP scores in the case study area.
Case Studies
In order to evaluate the individual and combined eects of
POEMS and TMS on system level MEP, a series of four
scenarios was evaluated in the SUMO model of Fort Collins,
Colorado. e scenarios were dened as follows:
Case Study 1: EMS (baseline) + TMS (baseline)
Case Study 2: EMS + Optimized TMS
Case Study 3: TMS + Optimized EMS (POEMS)
Case Study 4: Optimized EMS (POEMS) +
Optimized TMS
In each case, vehicle trajectories were generated using
SUMO and then EMS was applied post-hoc to those trajecto-
ries. Because EMS does not aect vehicle trajectories and TMS
does not use vehicle fuel economy as an input, EMS does not
need to beapplied in-loop with TMS.
Case Study 1:
In the rst case study the baseline EMS and baseline TMS
using validated Autonomie and SUMO, models are developed,
and corresponding baseline fuel economy and travel time
are calculated.
To calculate the baseline travel time from SUMO,
abstracted drive cycles including key parameters are fed along
with trac volume, trac network, vehicle model, and route
information. e key parameters from drive cycle abstraction
include acceleration rate, maximum speed, deceleration rate,
and emergency deceleration rate.
Aer calculating these parameters from real-life data, the
acceleration rate is 1.9 m/s2, deceleration rate is 2 m/s2, and
emergency deceleration rate is 6 m/s2. By comparing dierent
car following models in SUMO, EIDM (Extended Intelligent
Driver Model) is selected to represent the human driver
behavior because the travel time and lane change behavior
are most close to real-life data. Ta ble 4 shows the results
comparison from SUMO using EIDM and real-life data
driving the same route of the network.
Case Study 2:
e second case study considers the eect of optimized
TMS (trac signal time optimization) along with baseline
EMS on travel time and fuel economy values respectively.
Case Study 3:
e third case study considers the eect of baseline TMS
along with optimized EMS (POEMS) on travel time and fuel
economy values respectively.
Case Study 4:
e fourth and last case study considers the eect of opti-
mized TMS (trac signal time optimization) along with opti-
mized EMS (POEMS) on travel time and fuel economy
values respectively.
TAB LE 4 Comparison results from SUMO model (EIDM) and
real life data.
Parameters Real-Life Data EIDM
Travel time (s) 770 737
Travel distance(m) 6480 6506
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

7
MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL
ResultsandDiscussion
Figure 4 shows a comparison of the optimized results based
on running vehicles at each simulation time point (a), average
travel time vs simulation time step (b), and average speed vs
time step (c) with validated simulation results (case study 1
and 2).
Case Study MEP Impacts
For the four case studies outlined, weevaluate MEP impacts
to the area immediately surrounding the driving route (shown
in Figure 1), which is dened as a 250m2 buer bounding box
around the driving route. Table 5 presents the MEP impacts
for all four cases (case 1 being the baseline), and Figure 5
displays the spatial impact of each case to MEP at a 250m2
grid resolution. Improvements are observed to bethe strongest
along the southern part of the case study area along Prospect
Rd in all three optimal scenarios. While trac signal corridor
optimization in the optimal TMS only case improves travel
speeds and fuel economies along the route (as opposed to only
fuel economies and no eective impact on overall speed in
optimal EMS only), the optimal TMS only case shows the
smallest improvement. While the TMS optimization general ly
optimizes travel speeds on most of the links (up to 8 m/s
improvement over baseline and a lin k-length weighted average
of 20%), travel speeds on some links are sacriced for greater
corridor improvements (e.g., some link segments on College
Ave have slightly decreased speeds around -1 m/s from
baseline). is could be construed as a more realistic case
(where speeds on some links are improved at the cost of
reduced speeds on others) as opposed to the optimal EMS case,
where the network eciency (i.e., travel speeds on the links)
is assumed to bepractically immune to the eciencies brought
about by improved fuel economy. Further investigation into
the tradeo s in EMS only and TMS only scenar ios is warranted.
e cumulative improvement from combined EMS and
TMS optimization also results in MEP improvements that
surpass the arithmetic summation of MEP impacts EMS only
TAB LE 5 MEP impacts for four case studies evaluated. These impacts consider MEP scores only in a 250m2 buer around the
route in Fort Collins, Colorado. For these results, the road-miles impacted was 8.2% of all road-miles in the study area (i.e., most
links in the area are not modified with improvements). TT = Travel Time.
Case Study
Mean FE and TT
Improvements Drive MEP
Relative MEP
Improvement
1 Baseline - 29.36 -
2 Optimal TMS only 16.1% FE; 20% TT 29.44 0.28%
3 Optimal EMS only 17.7% FE; 0% TT 29.60 0.82%
4 Optimal EMS+TMS 43.2% FE; 20% TT 29.82 1.57%
FIGURE 5 comparison of running time (a), Travel time (b)
and Average speed (c) vs. time for case studies with baseline
TMS (case study 1 and 3) and optimized TMS (case study 2
and 4)
FIGURE 4 SUMO Network.
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL 8
and TMS only optimization. e increase in cumulative gains
is derived primarily from the increase in FE for the optimal
EMS+TMS case. is shows that the combination of optimal
EMS and TMS in concert is best for increasing the quality of
driving accessibility in terms of MEP. It is oen the case that
EMS and TMS optimizations are conducted and implemented
in isolation as the stakeholders interested in these improve-
ments are fairly disjoint. While the trac operations commu-
nity is primarily responsible for nding and implementing
TMS improvements, EMS improvements come from vehicle
manufacturers. e result here corroborates that idiom “e
whole is greater than the sum of the parts,” and that it would
bebenecial for both vehicle and trac system communities
to come together for delivering the greatest (MEP) benets to
the travelers.
It is important to note here that the case studies eva luated
show realistic impacts that are achievable for only improve-
ments on a small corridor (roughly 2km2 area with only 8.2%
of road-miles simulating optimization). Wetested a scenario
that translates the average improvements by road class across
the whole city of Fort Collins, CO, and found that an optimal
TMS and EMS case that applies to 7.2% of road-miles across
the city improved the city-wide MEP score by 18.5%. is
shows that EMS and TMS optimization when applied at scale
may result in signicant improvement of the mobility and
energy efficiency of a city's transportation system. Full
network optimizations need to be tested to corroborate
this claim.
SummaryandConclusions
In a connected world, vehicles and infrastructure can and
must bedeveloped in a synergistic manner in order to enable
the greatest eciency for a given transportation system as a
whole. Using previously developed optima l EMS and optimal
TMS methods individually and in tandem, this study demon-
strates that substantia l additional eet level eciency benets
may beattained by the application of both simultaneously
over the combined eects of each individually. FE improve-
ments of 16.1% and 17.7% over baseline were obtained respec-
tively from the application of optimal TMS and EMS indi-
vidually where a 43.2% improvement over baseline was
obtained from their simultaneous application. ese FE
improvements, combined with travel time improvements
from optimal TMS, resulted in MEP improvements of 0.28%
and 0.82% for optimal TMS and optimal EMS respectively
and 1.57% for the application of both simultaneously. In both
FE and MEP terms the improvements from the application
of both optimal EMS and optimal TMS were greater than
the sum of the individual improvements. e results of this
study illustrate the fundamental interconnectedness of
vehicle level and infrastructure level energy optimization
and underscore the importance of the benets which can
be attained through connectivity. Overall, by finding
evidence of a positive synergistic relationship between vehicle
and transportation system level optimal controls this study
lays the groundwork for a new direction of research in collab-
orative development between transportation stakeholders to
optimize system level eciency.
References
1. “Energy Ecient Mobility Systems FY 2020 Annual Progress
Report,” October 2021, https://www.energy.gov/eere/
vehicles/articles/energy-ecient-mobility-systems-fy-2020-
annua l-progress-report.
2. EPA, US and OAR, “Inventory of U.S. Greenhouse Gas
Emissions and Sinks: 1990-2016,” 2018.
3. Motallebiaraghi, F., Rabinowitz, A., Jathar, S., Fong, A. et al.,
“High-Fidelity Modeling of Light-Duty Vehicle Emission
and Fuel Economy Using Deep Neural Networks,” SAE
Technical Paper 2021-01-0181 (2 021), d oi:10.4 271/2021-01-
0181.
4. Gaikwad, T., Rabinowitz, A., Motallebiaraghi, F., Bradley, T.
et al., “Vehicle Velocity Prediction Using Articial Neural
Network and Eect of Real World Signals on Prediction
Window,” SAE Technical Paper 2020-01-0729 (2020),
doi:10.4271/2020-01-0729.
5. Tunnell, J., Asher, Z.D., Pasricha, S., and Bradley, T.H.,
“Toward Improving Vehicle Fuel Economy with ADAS,” SAE
Technical Paper 2018- 01-0593 (2 018), doi:10.4271/2018- 01-
0593.
FIGURE 6 MEP scores at 250m2 resolution for the four
scenarios (a-d) simulated. The driving route which had
modified speed and fuel economies is labeled in (a) and shown
as a black line overlay.
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

9
MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL
6. Asher, Z.D., Patil, A.A., Wif vat, V.T., Frank, A.A. et al.,
“Identication and Review of the Research Gaps Preventing
a Realization of Optimal Energy Management Strategies in
Veh icles,” SAE Int. J. Alt. Power. 8, no. 2 (2019): 133-150,
doi:10.4271/08-08-02-0009.
7. Mohd Zulkei, M.A., Zheng, J., Sun, Z., and Liu, H.X.,
“Hybrid Powertrain Optimization with Trajectory
Prediction Based on Inter-Vehicle-Communication and
Vehicle-Infrastructure-Integration,” Transp. Res. Part C:
Emerg. Technol. 45 (2014): 41-63.
8. Luo, Y., Chen, T., Zhang, S., and Li, K., “Intelligent Hybrid
Electric Vehicle ACC With Coordinated Control of Tracking
Ability, Fuel Economy, and Ride Comfort,” IEEE Trans.
Intell . Transp. Syst. 16, no. 4 (2015): 2303-2308.
9. Wang, Q. and Ayalew, B., “A Probabilistic Framework for
Tracking the Formation and Evolution of Multi-Vehicle
Groups in Public Trac in the Presence of Observation
Uncertainties,” IEEE Trans. Intell. Transp. Syst. 19, no. 2
(2018): 560-571.
10. Mahmoud, Y.H., Brown, N.E., Motallebiaraghi, F., Koelling,
M. et al., “Autonomous Eco-Driving with Trac Light and
Lead Vehicle Constraints: An Application of Best
Constrained Interpolation,” IFAC-PapersOnLine 54, no. 10
(2021): 45-50.
11. Goberville, N., Zoardar, M.M., Rojas, J., Brown, N. et al.,
“Techno-Economic Analysis of Fixed-Route Autonomous
and Electric Shuttles,” SAE Technical Paper 2021-01-0061
(2021), d oi:10.4271/2021- 01- 0 061.
12. Martinez, C.M., Hu, X., Cao, D., Velenis, E. et al., “Energ y
Management in Plug-In Hybrid Electric Vehicles: Recent
Progress and a Connected Vehicles Perspective,” IEEE Trans.
Veh. Technol. 66, no. 6 (2017): 4534-4549.
13. Uebel, S., Murgovski, N., Tempelhahn, C., and Bäker, B.,
“Optimal Energy Management and Velocity Control of
Hybrid Electric Vehicles,” IEEE Trans. Veh. Technol. 67, no. 1
(2 018): 327- 337.
14. Lin, C.-C., Kang, J.-M., Grizzle, J.W., and Peng, H., “Energy
Management Strategy for a Parallel Hybrid Electric Truck,”
in Proceedings of the 2001 American Control Conference.
(Cat. No.01CH37148), 2001, doi:10.1109/acc .2001.946337.
15. Asher, Z.D., Trinko, D.A., Payne, J.D., Geller, B.M. et al.,
“Real-Time Implementation of Optimal Energy Management
in Hybrid Electric Vehicles: Globally Optimal Control of
Acceleration Events,” J. Dyn. Syst. Meas. Control 142, no. 8
(2020), doi:10.1115/1.40 46477.
16. Serrao, L., Onori, S., and Rizzoni, G., “ECMS as a Realization
of Pontryagin’s Minimum Principle for HEV Control,” 2009
American Control, 2009.
17. Wu, J., Ruan, J., Zhang, N., and Walker, P.D., “An Optimized
Real-Time Energy Management Strategy for the Power-Split
Hybrid Electric Vehicles,” IEEE Trans. Control Syst. Technol.
27, no. 3 (2019): 1194-1202.
18. Liang, C., Xu, X., Wang, F., and Zhou, Z., “Coordinated
Control Strategy for Mode Transition of DM-PHEV Based
on MLD,” Nonlinear Dyn. (2 021).
19. Rabinowitz, A., Araghi, F.M., Gaikwad, T., Asher, Z.D. et al.,
“Development and Evaluation of Velocity Predictive
Optimal Energy Management Strategies in Intelligent and
Connected Hybrid Electric Vehicles,” Energies 14, no. 18
(2021): 5713 .
20. Paganelli, G., Ercole, G., Brahma, A., Guezennec, Y. et al.,
“General Super visory Control Policy for the Energy
Optimization of Charge-Sustaining Hybrid Electric
Veh icles,” JSAE Review 22, no. 4 (2001): 511-518.
21. Patil, A.A., Motallebiaraghi, F., Meyer, R., and Asher, Z.D.,
“Comparison of Optimal Energy Management Strategies
Using Dynamic Programming, Model Predictive Control,
and Constant Velocity Prediction,” SAE Technical Paper
2020-01-5071 (2020), doi:10.4271/2020-01-5071.
22. Bender, F.A. and Kaszynski, M., “Drive Cycle Prediction and
Energy Management Optimization for Hybrid Hydraulic
Veh icles,” IEEE Transactions on Vehicular Technology 62, no.
8 (2013): 3581-3592.
23. Donateo, T., Pacella, D., and Laforgia, D., “Development of
an Energy Management Strategy for Plug-In Series Hybrid
Electric Vehicle Based on the Prediction of the Future
Driving Cycles by ICT Technologies and Optimized Maps,”
SAE Technical Paper 2011- 01- 0892 (2011), do i :10.4271/2 011-
01-0 892 .
24. Sun, C., Moura, S.J., Hu, X., Hedrick, J.K. et al., “Dynamic
Trac Feedback Data Enabled Energy Management in Plug-
In Hybrid Electric Vehicles,” IEEE Trans. Control Syst.
Technol. 23, no. 3 (2015): 1075-1086.
25. Sun, C., Hu, X., Moura, S.J., and Sun, F., “Velocity Predictors
for Predictive Energy Management in Hybrid Electric
Veh icles,” IEEE Trans. Control Syst. Technol. 23, no. 3 (2015):
1197-1204 .
26. Pandit, K., Ghosal, D., and Zhang, H.M., “Adaptive
TracSignal Control with Vehicular Ad Hoc Networks,”
IEEE Transactions on Vehicular Technology 62, no. 4 (2013):
1459-14 71.
27. Gartner, N.H., Stamatiadis, C., and Tarno, P.J.,
“Development of Advanced Trac Signal Control Strategies
for Intelligent Transportation Systems: Multilevel Design,”
Transp. Res. Rec. (1995).
28. Sheikh, M.S., Liang, J., and Wang, W., “An Improved
Automatic Trac Incident Detection Technique Using a
Vehicle to Infrastructure Communication,” Journal of
Advanced Transportation 2020 (2020),
doi:10.1155/2020/9139074.
29. Chang, T.H. and Lin, J.T., “Optimal Signal Timing for an
Oversaturated Intersection,” Trans. Res. Part B:
Methodol. (2000).
30. He, Y., Rong, Y., Liu, Z., and Du, S., “Trac Inuence Degree
of Urban Trac Emergency Based on Water Wave Principle,”
Journal of Transportation Systems Engineering and
Information Technology (2017), doi:10.16097/j.cn ki.10 09-
6744.2017.05.021.
31. Han, E., Lee, H.P., Park, S., So, J. et al., “Optimal Signal
Control Algorithm for Signalized Intersections under a V2I
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL 10
Communication Environment,” Journal of Advanced
Transportation 2019 (2019), doi:10.1155/2019/6039741.
32. Wang, J., Hang, J., and Zhou, X., “Signal Timing
Optimization Model for Intersections in Trac Incidents,”
Journal of Advanced Transportation 2020 (2020),
doi:10.1155/2 02 0/1081365.
33. Garikapati, V., Young, S., and Hou, Y., “Measuring
Fundamental Improvements in Sustainable Urban Mobility:
e Mobility-Energ y Productivity Metric,” in International
Conference on Transportation and Development 2019, 2019,
doi:10.1061/9780784 482582.010.
34. Hou, Y., Garikapati, V., Nag, A., Young, S.E. et al., “Novel
and Practical Method to Quantify the Quality of Mobility:
Mobility Energy Productivity Metric,” Transp. Res. Rec.
2673, no. 10 (2019): 141-152.
35. Guo, L., Gao, B., Gao, Y., and Chen, H., “Optimal Energy
Management for HEVs in Eco-Driving Applications Using
Bi-Level MPC,” IEEE Trans. Intell. Transp. Syst. 18, no. 8
(2017): 2153-2162.
36. Li, M., Wu, X., He, X., Yu, G. et al., “An Eco-Driving System
for Electric Vehicles with Signal Control under V2X
Environment,” Transp. Res. Part C: Emerg. Technol. 93
(2018): 335-350.
37. Tulpule, P., Marano, V., and Rizzoni, G., “Eect of Trac,
Road and Weather Information on PHEV Energ y
Management,” 2011.
38. He, Y., Chowdhury, M., Ma, Y., and Pisu, P., “Merging
Mobility and Energy Vision with Hybrid Electric Vehicles
and Vehicle Infrastructure Integration,” Energy Policy 41
(2012): 599-609.
39. Tian, H., Wang, X., Lu, Z., Huang, Y. et al., “Adaptive Fuzzy
Logic Energ y Management Strategy Based on Reasonable
SOC Reference Curve for Online Control of Plug-In Hybrid
Electric City Bus,” IEEE Trans. Intell. Transp. Syst. 19, no. 5
(2018): 16 07-1617.
40. Shim, Y. and Mollo, C., “A Reinforcement Learning
Algorithm for Speed Optimization and Optimal Energy
Management of Advanced Driver Assistance Systems and
Connected Vehicles,” SAE Int. J. Commer. Veh. 14, no. 3
(2021): 289-302, doi:10.4271/02-14-03-0023.
41. Sun, C., Shen, X., and Moura, S., “Robust Optimal ECO-
Driving Control with Uncertain Trac Signal Timing,” in
2018 Annual American Control Conference (ACC), 5548-
5553, 2018.
42. Gaikwad, T.D., Asher, Z.D., Liu, K., Huang, M. et al.,
“Vehicle Velocity Prediction and Energy Management
Strategy Part 2: Integration of Machine Learning Vehicle
Velocity Prediction with Optimal Energy Management
to…,” 2019.
43. Serrao, L., Onori, S., and Rizzoni, G., “ECMS as a Realization
of Pontryagin’s Minimum Principle for HEV Control,” in
2009 American Control Conference, 3964-3969, 200 9.
44. Musardo, C., Rizzoni, G., Guezennec, Y., and Staccia, B.,
“A-ECMS: An Adaptive Algorithm for Hybrid Electric
Vehicle Energy Management,” Eur. J. Control 11, no. 4 (2005):
509-524.
45. Sun, C., He, H., and Sun, F., “e Role of Velocity
Forecasting in Adaptive-ECMS for Hybrid Electric Vehicles,”
Energy Procedia 75 (2015): 1907-1912.
46. “Downloadable Dynamometer Database,” October 2021,
https://www.anl.gov/es/downloadable-dynamometer-
database.
47. Zubillaga, D., Cruz, G., Aguilar, L.D., Zapotécatl, J. et al.,
“Measuring the Complexity of Self-Organizing Trac
Lights,” Entropy 16, no. 5 (2014): 2384-2407.
48. Akcelik, R., “Extension of the Highway Capacity Manual
Progression Factor Method for Platooned Arrivals,” 1995.
49. A kçelik, R. and Rouphail, N.M., “Overow Queues and
Delays with Random and Platooned Arrivals at Signalized
Inters ect ions,” J. Adv. Transp. 28, no. 3 (1994): 227-251.
50. Cowan, R., “An Improved Model for Signalised Intersections
with Vehicle-Actuated Control,” J. Appl. Probab. 15, no. 2
(1978): 384 -396.
51. “National Household Travel Survey,” October 2021, https://
nhts.ornl.gov/.
Contact Information
Farhang Motallebiaraghi
Western Michigan University, Kalamazoo, MI 49008 USA
farhang.motallebiaraghi@wmich.edu
Zachary D. Asher, Ph.D.
Western Michigan University, Kalamazoo, MI 49008 USA
zach.asher@wmich.edu
Acknowledgment
is material is based upon work supported by the U.S.
Department of Energy’s Office of Energy Efficiency and
Renewable Energy (EERE). e specic organization over-
seeing this report is the Vehicle Technologies Oce under
award number DE-EE0008468.
Definitions/Abbreviations
ADAS - Advanced Driver Assistance Systems
ANL D3 - Argonne National Lab’s Downloadable
Dynamometer Database
ANN - Articial Neural Network
CAV - Connected and Automated Vehicle
DP - Dynamic Programming
DPS - Dynamic Phase Selection
ECMS - Equivalent Consumption Minimization Strategy
EIDM - Extended Intelligent Driver Model
EMS - Energy management System
EPA - Environmental Protection Agency
FE - Fuel Economy
GHG - Greenhouse Gas
HEV - Hybrid Electric Vehicle
ICE - Internal Combustion Engine
ITS - Intelligent Trac System
LSTM - Long Short-Term Memory
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022

© 2022 SAE I nternational. A ll rights reser ved. No part of thi s publication may b e reproduced, s tored in a retrieval sys tem, or transmit ted, in any form or by any m eans,
electronic, mechanical, ph otocopying, recording, or otherwise, without the prior written permission of SAE International .
Positions and opinions advanced in this work are those of the author(s) and not necessarily those of SAE I nternational. Responsibility for the content of the work lies
solely with the author(s).
ISSN 0148-7191
11
MOBILITY ENERGY PRODUCTIVITY EVALUATION OF PREDICTION-BASED VEHICLE POWERTRAIN CONTROL
MaxDur - Maximum Duration
MEP - Mobility Energy Productivity
ML - Machine Learning
MPC - Model Predictive Control
NHTS - National Household Travel Survey
NREL - National Renewable Energy Laboratory
PHEV - Plug-in Hybrid Electric vehicle
POEMS - Predictive Optimal Energy Management Strategy
QLD - Queue Length Dissipation
SOC - State of Charge
SUMO - Simulation of Urban Mobility
TMS - Trac Management System
TT - Travel Time
UDDS - Urban Dynamometer Driving Schedule
V2I - Vehicle to Infrastructure
V2V - Vehicle to Vehicle
V2X - Vehicle to Everything
Downloaded from SAE International by Farhang Motallebiaraghi, Monday, March 28, 2022