Electric Train Energy Consumption Modeling

Jinghui Wanga , Hesham A. Rakhab,∗

a
Center for Sustainable Mobility,Virginia Tech Transportation Institute,3500 Transportation Research
Drive, Blacksburg, VA 24061,USA
b
Center for Sustainable Mobility,Virginia Tech Transportation Institute,3500 Transportation Research
Drive, Blacksburg, VA 24061,USA

Abstract
The paper develops an electric train energy consumption modeling framework considering
instantaneous regenerative braking efficiency in support of a rail simulation system. The
model is calibrated with data from Portland, Oregon using an unconstrained non-linear
optimization procedure, and validated using data from Chicago, Illinois by comparing
model predictions against the National Transit Database (NTD) estimates. The results
demonstrate that regenerative braking efficiency varies as an exponential function of the
deceleration level, rather than an average constant as assumed in previous studies. The
model predictions are demonstrated to be consistent with the NTD estimates, producing
a predicted error of 1.87% and -2.31%. The paper demonstrates that energy recovery
reduces the overall power consumption by 20% for the tested Chicago route. Furthermore,
the paper demonstrates that the proposed modeling approach is able to capture energy
consumption differences associated with train, route and operational parameters, and thus
is applicable for project-level analysis. The model can be easily implemented in traffic
simulation software, used in smartphone applications and eco-transit programs given its
fast execution time and easy integration in complex frameworks.
Keywords: Electric train, Energy Consumption Model, Regenerative Braking Efficiency,
Rail Transit Simulation

1. Introduction
The transportation sector has become the major consumer of energy and producer of
greenhouse gas (GHG) emissions. Specifically, in 2014, transportation-related energy use
accounted for 27% of the total world primary energy consumption and produced 34% of

∗
Corresponding Author.
Email address: [email protected] (Hesham A. Rakha)

Preprint submitted to Journal of Applied Energy February 1, 2017

© 2017. This manuscript version is made available under the Elsevier user license
http://www.elsevier.com/open-access/userlicense/1.0/
GHG emissions (CO2 ) [1]. In the United States, the transportation sector is responsible
for 28% of the total U.S. energy use and 34% of CO2 production [2, 3]. The urban trans-
portation system has been deteriorating with regards to the environment given the higher
residential density and travel demand, growth in automobile ownership and worsening of
traffic conditions. Many cities, especially metropolis areas, are served by a mixture of
multiple traffic modes comprised of passenger cars, transit buses, trucks and rail transit.
Accordingly, to reduce city-wide energy consumption and GHG emissions, not only are
effective strategies required for each mode, but also integrated strategies considering the
interaction of these modes are required. To enable the efficient and cost-effective design
and testing of new strategies, a multi-modal simulation system is being developed in order
to offer a simulated testbed applicable to multiple traffic modes.
Developing the proposed system requires the design of three modules: passenger car
and transit vehicle modeling, subway system modeling, pedestrian and bicycle modeling.
This paper focuses on subway system electric rail energy consumption modeling. The
developed model will support the overall modeling framework by estimating rail-induced
energy consumption and GHG emissions, and designing and testing eco-friendly strategies
customized to urban rail transit systems, such as energy-efficient timetables [4, 5] and eco-
speed control [6, 7].

2. Literature review
Energy consumption models can be divided into two categories: forward models and
backward models. As demonstrated by [8], models that compute the tractive contribu-
tion required at the wheels and “work backward” towards the engine are called “backward
models”; alternatively, models that start from the engine and work with transmitted and re-
flected torque are called “forward models”. The use of forward models requires extensive
internal engine data. These models are very complex and characterized by slow execution
time and high computer memory. Backward models, however, achieve reliable evaluation
of vehicle energy consumption based on drive cycle and vehicle characteristic data, with-
out the need to input engine data. In addition, they are characterized by fast computational
times and low memory usage, and can be easily implemented in complex frameworks such
as simulation and intelligent transportation system (ITS) applications [9]. The backward
modeling approach is thus used to develop the proposed modeling framework.
The effort on the review of the literature was made in terms of introducing the details
of a subway system along with the system components relative to energy consumption
modeling, and the existing backward modeling approaches on rail energy consumption.

2
2.1. Subway system components for energy modeling
A subway, put simply, is a train and the tunnel through which the train runs. A subway
train consists of several connecting cars that contain durable seats as well as poles and
straps for people to hold on to when the train is full [10]. The trains, known as rolling
stock, are complex given that they include a traction and dynamics system that highly
impact energy consumption. For example, the traction system determines how the propul-
sive force is generated and provided to move a train forward; and the dynamics system
determines how a train is accelerated or decelerated thus affecting train transient behavior
that highly impacts instantaneous energy consumption. Also, the brake system determines
whether the braking power is regenerated to be used or wasted as heat. For regenerative
braking, the brake energy can be recovered by converting kinetic energy into a form that
can be either immediately used or stored until needed; however, other brake systems, such
as dynamic braking, dissipate electric energy as heat rather than using it. Other train char-
acteristics, such as car empty weight, number of axles per rail car and drag coefficient
significantly affect the forces acting on a train and thus are also important parameters in
energy modeling.
In addition to the train itself, there are several other rail system components, such as
track infrastructure and passenger loading, affecting the tractive/braking forces acting on
the train. For instance, a good condition track (good rails and cross ties) decreases the
starting tractive effort [11]; while a track with steep grades and large curvature results in
high resistance forces [11, 12]; and passenger loading affects the total railcar weight and
thus acting forces. These factors should also be incorporated into the energy modeling
framework.

2.2. Existing rail energy modeling approaches

The most widely available measures for rail (either electric or diesel-electric train) en-
ergy consumption are those estimated on an annual gross average basis. Specifically, Eq.
(1)-(3) present the modeling approach, where Ep , Es and Ev are the energy consumed per
passenger kilometer (kW h/P · km), per seating kilometer (kW h/S · km) and per vehi-
cle kilometer (kW h/V · km), respectively; E is the annual energy consumption of a rail
transit system in kW h; Mp , Ms and Mv are total passenger kilometers, seating kilometers
and vehicle kilometers, respectively; C is the train seating capacity and β is the line loss
factor associated with the train transmission system. The parameters in the models are
readily available from the National Transit Database (NTD) [13–15]. Despite the effort-
less acquisition of the aggregated measurements, they are not capable of representing the
differences in energy consumption associated with route and vehicle characteristics, pas-
senger loading, speed profiles and weather and track conditions, and thereby not suitable

3
for project-level analysis.

E
Ep = (1)
Mp × β

E
Es = (2)
Ms × C × β

E
Ev = (3)
Mv × β
Research efforts have thus focused on developing a modeling framework sensitive to
the aforementioned system characteristics. An early study conducted by Mittal [16] pro-
posed an analytical method to estimate energy consumption sensitive to speed, train con-
figuration and passenger load. However, the method used average speed for energy predic-
tion without considering speed fluctuation. The model also did not incorporate an energy
regeneration module. Some of the state-of-the-art models [17–22] considered an average
constant regenerative braking energy efficiently that mainly depended on the train’s av-
erage speed. The major limitation is that these models cannot capture vehicle transient
behavior and model energy regeneration at a microscopic level. Although some of these
simplified models have been used to develop energy-optimized strategies [23–28], the
validity of the resulting strategies is questionable given the models’ inadequacy in instan-
taneous energy prediction. The National Cooperative Rail Research Program (NCRRP)
[29] designed a passenger rail simulation framework which incorporated route and train
characteristics, speed, passenger load and regenerative braking into the energy modeling
practice. The framework, however, is an excel-based tool and cannot be implemented in
more complex frameworks, such as traffic simulation software, smartphone eco-driving
and eco-routing systems. Furthermore, the energy prediction within the framework also
considers average speed and a constant regenerative efficiency. A recent study [12] initi-
ated a bottom-up modeling framework sensitive to acceleration behavior by incorporating
second-by-second speed profiles. Nonetheless, the model cannot generate instantaneous
energy regeneration because it assumes a constant regenerative efficiency. The model is
also an excel-based tool and thus does not allow for integration in complex frameworks.
Other models, such as [30–32], are also not suitable for ITS applications due to their com-
plexity in model specification.
To the authors’ best of knowledge, although there have been numerous studies on mod-
eling rail electric consumption, these studies were of limited application especially for road

4
electric vehicles ([33–37]). Specifically, they either cannot model train transient behavior
or fail to capture energy regeneration at a microscopic level or are not simple enough to
be implemented within complex systems (e.g. traffic simulation software and smartphone
applications). The paper attempts to fill this void and, for the first time, relates energy
regeneration with the instantaneous deceleration level in rail transit energy modeling.

3. Modeling framework
The proposed modeling framework characterizes the energy prediction as two piece-
wise functions, as demonstrated in Eq.(4) (energy consumption) and Eq.(5) (energy regen-
eration). The description of the model parameters is summarized in Table A1 in Appendix
A. Basically, the energy is computed on a second-by-second basis. When the train is in
traction mode, the energy flows from the electricity power system to the wheels with the
power at the wheels being positive (P > 0). Alternatively, when the train is in regenerative
braking mode, the energy flows from the wheels back to the power system and the power
at the wheels is negative (P < 0). Noteworthy here is that, to compute regenerated energy
(ECre ), only negative power is considered.
(α01 × β1 + α02 × β2 ) in Eq.(4) refers to the head-end power (HEP) in which β1 and
β2 are dummy variables equal to either 0 or 1. α01 , in most cases, is applied to HEP
(β1 = 1 and β2 = 0) except whan a train is about to start moving and waiting at the initial
route station where only a small fraction of HEP (α02 ) is applied (β1 = 0 and β2 = 1).
This accounts for the fact that trains only keep the ventilation system and lights on while
waiting to load passengers before a trip begins and thus only consumes a small fraction of
HEP.


α01 × β1 + α02 × β2 + P (t), ∀P (t) > 0
EC(t) = (4)
α01 × β1 + α02 × β2 , ∀P (t) ≤ 0


P (t) × ηrb (t), ∀P (t) < 0
ECre (t) = (5)
0, ∀P (t) ≥ 0

The average energy consumption rate for an entire trip is then estimated by summing
the instantaneous energy predictions and then dividing by the trip length, as illustrated in
Eq. (6), with d being the trip length (km).

  
t [EC(t)
+ ECre (t)]
ECd kW h/V · km = (6)
d

5
3.1. Tractive power and tractive effort
Tractive power is computed using Eq.(7) in which u is the instantaneous speed (km/h),
0.746 is used to convert the power from horsepower to kilowatt. F is the tractive effort as
2
formulated in Eq.(8) [11]. The first four terms, (0.6 + w20p + 0.01u(t)
1.61
+ K(u(t)/1.61)
wp n p
), in the
model are the modified Davis equation [38], referring to the train resistance comprised of
rolling, journal, track, flange and aerodynamic resistance. wp is the railcar weight per axle
(ton), including passenger weight (an average of 68 kg is assumed for each passenger in
this study). The Davis equation was tested through a large amount of field experiments
[38–40]. In addition, as demonstrated by [11, 12, 41], only positive gradient contributes to
the grade resistance with an increase of 20 lbs/ton (0.01 N/kg) per percentage grade. It
should be noted that curve resistance has been converted to the equivalent grade resistance
by assuming that unit resistance of a 1◦ curve is the same as the resistance that a 0.04%
grade would offer [11]. The last term in the bracket is the force exerted for acceleration
or braking. M is the average weight of the moving train (ton), including the train curb
weight and total passenger weight. 4.4482 is used to convert the tractive effort from lbs to
N.

Fu
P = × 0.746 (7)
375 × 1.61

20 0.01u(t) K(u(t)/1.61)2

F (t) = (0.6 + + + + 20θ)
wp 1.61 w p np
u(t)2 − u(t − 1)2

+ 70 × M × 4.4482 (8)
8.4 × L
3.2. Starting tractive effort
It is worth noting that the tractive force in Eq.(8) only addresses the effort exerted to
move the train while in motion. However, a different tractive force is needed to move a
train from a complete stop. The starting tractive effort typically consists of the grade resis-
tance, bearing resistance, track resistance, weather resistance, and the resistance resulting
from poor track conditions. When a train is starting from a complete stop, the tractive
effort is estimated as the sum of these resistance forces rather than using Eq. (8).
The grade resistance can be estimated as set forth. The typical values of other re-
sistance forces were suggested by [12]. Specifically, the bearing resistance is 10 lb/ton
(0.005 N/kg) at 122 ◦ C (50 ◦ F), and increases by 0.1 lb/ton (0.0005 N/kg) for 1 ◦ F de-
crease below 50 ◦ F and decreases by 0.1 lb/ton (0.00005 N/kg) for 1 ◦ F increase above
50 ◦ F. Track resistance depends on track type. There is no resistance force for 130 lb (59

6
kg) rail, and 1 lb/ton (0.0005 N/kg) resistance for 115 lb (52 kg) rail and 2 lb/ton (0.001
N/kg) for 100 lb (45 kg) rail. Weather resistance is affected by the humidity of the rail.
Basically, there is no weather resistance for dry rail, while the wet rail produces a 2 lb/ton
(0.001 N/kg) resistance. The resistance of icy or snowy rail goes up to 10 lb/ton (0.005
N/kg). The resistance relative to track conditions is 2 lb/ton (0.001 N/kg) for poor rails
and fair cross ties, and 7 lb/ton (0.0035 N/kg) for poor rails and poor cross ties. There is
no such resistance if both rails and cross ties are in good condition.

3.3. Regenerative braking efficiency

Regenerative braking efficiency accounts for the portion of the total braking energy
available for recovering. It determines the amount of the energy recovered by a regen-
erative braking system. The trains operating on an urban rail transit system frequently
accelerate and decelerate given the short distance between two stations, so that a large
amount of energy may potentially be recovered. Failing to account for energy recovery
may thus result in large deviation in energy prediction for trains with regenerative braking.
The regenerative efficiency (ηre ) was characterized as an exponential function of the
deceleration level for electric vehicles [8, 42], as formulated in Eq.(9). The model demon-
strates that higher deceleration levels result in larger regenerative efficiency and thus more
energy recovery. This functional form was thus used in the proposed model to compute
the train regenerative energy. The calibration of the model parameter (α) is presented in
Section 4.

 1
α , ∀a(t) < 0
ηre (t) = e |a(t)| (9)
0, ∀a(t) ≥ 0

4. Model calibration
Having introduced the modeling framework, the next step is to calibrate the proposed
model, which requires only one parameter to be calibrated.
The data required for calibration were classified into four categories: train informa-
tion, travel activity data, route characteristics, and the information required to estimate
the starting tractive effort. The data were provided by researchers at Georgia Tech who
requested the data from the Tri-County Metropolitan Transportation District of Oregon
(TriMet), the public agency that operates mass transit in the Portland Metropolitan area.
TriMet responded with the information for the Metropolitan Area Express (MAX) Blue
Line.
The MAX light rail trains are powered by built-in-place electric substations (ESS) lo-
cated along the system route using the overhead contact line ([43]) with a nominal Voltage

7
 Fig. 1: A simplified representation of the MAX light rail traction system (Source: [44])

of 825 V DC. The system under study is similar to that investigated by [44, 45]. As il-
lustrated in Fig. 1, the electric power is transferred from ESS to trains through the contact
line. The brake system of the MAX trains is a blending of regenerative braking and fric-
tion braking. The regenerative braking is the primary method of braking used when the
train is going faster than 3 mph (4.8 km/h), and the friction braking is also applied when
the speed of the train is slower than 4,8 km/h to ensure sufficient power supply for emer-
gency stopping. In addition, when the train is in braking mode, the electric motors operate
like generators, taking the forward motion of the train and converting it into electricity.
The regenerated electricity is then partly sent to other trains and partly stored in lineside
storage systems for later usage, which significantly improves the system power efficiency.
For example, as demonstrated by Fig. 1, the braking power from train A is partly sent to
train B and partly stored in the storage system around ESS 2.
A typical train model (SD 660), as shown in Fig. 2, manufactured by Siemens, was
tested in the field by TriMet to construct the calibration dataset. The test vehicle had two
connecting cars each of which provided with 64 seats. The specific information of the
testing rolling stock is illustrated in Table 1, including the empty railcar weight, number
of axles per rail car, drag coefficient, seating capacity, passenger loading, number of cars
per train, and HEP. The HEP was simplified to three operational levels: normal, high and
maximum. The normal level operates at one-third of the maximum load (25 kW ), and the

8
 Fig. 2: Portland MAX light rail vehicle

high level functions at two-thirds followed by the maximum level that operates at full load.
It should be noted that the Chicago train in the table was used for validation purposes.
The characteristics of the testing route are provided in Table B1 of Appendix B in
terms of station name, milepost, elevation and gradient profiles (the curvature has been
converted to its equivalent grade as mentioned in section 3.1). The trip starts from the
Hatfield Government Center Station and ends at Cleveland station, covering an entire trip
of the line with a total distance of 32.4 miles (52.16 km). The vertical layout of the
section demonstrates a combination of upgrade (up to 4.26%) and downgrade (up to -
4.46%) sections.
The information required to estimate the starting tractive effort, as illustrated in Ta-
ble 2, consists of weather condition, ambient temperature, track type and conditions. The
test runnings in Portland and Chicago were both conducted on 115-lb good rails with good
cross ties at normal temperature on sunny days. The test in Portland delivered the MAX
Blue Line driving cycle with a maximum operational speed of 88 km/h and a total du-
ration of around 6000 s, as illustrated in Fig. 3. For each between-station running, the
train, in general, was first accelerated from a complete stop at one station to a specified
target speed, then cruised at the target speed for a spell, and decelerated until achieving a
complete stop at the next station.
The exponential feature of the regenerative efficiency in Eq. (9) results in a non-

9
 Table 1: Test train characteristics

Values
Parameter MAX blue Chicago Brown
Weight of empty car (ton) 54.5 27.15
Number of axles per car 6 4
Drag coefficient 0.07 0.07
Seating capacity per car 64 49
Percentage loading (peak period) na 87.5%
Percentage loading (off-peak period) na 25%
Daily percentage loading 43% 45%
Number of cars per train (peak period) 2 6
Number of cars per train (off-peak period) 2 4
Maximum HEP per car (kW ) 25 25
HEP operating level Normal Normal

Table 2: Parameter set for starting tractive effort

Values
Parameter MAX Blue Chicago Brown
Ambient temperature (◦ F) 74 74
Weather condition Dry Dry
Track type 115 lb (52 kg) rail 115 lb (52 kg) rail
Track condition Good rails and crossties Good rails and crossties

10
 90

80

70

60
Speed (km/h)

50

40

30

20

10

0
0 1000 2000 3000 4000 5000 6000
Time (s)

Fig. 3: The driving cycle for the MAX Blue Line

linear energy consumption model, and thus the calibration procedure was formulated as
an unconstrained non-linear optimization problem as shown in Eq. (10), where D is the
model prediction difference relative to the NTD estimates, and ECN T D is the NTD average
energy rate. The NTD energy rate was used to calibrate the model because the field energy
consumption was unavailable at the moment of model development. The data in the NTD
2011 demonstrated an average energy consumption of 13.57 kW h/V · mi (8.48 kW h/V ·
km). The model was calibrated by varying the model parameter (α) value to achieve the
minimum prediction error.

min D = (ECd − ECN T D )2 (10)

The calibration result generates the optimum model parameter of 0.65 (α = 0.65).
The resulting regenerative efficiency, as demonstrated in Fig. 4, exponentially decays with
a decrease in the deceleration level. In particular, the decay becomes dramatic when the
deceleration level is less than 2 m/s2 .

5. Model validation
The model was validated using the data from the Chicago heavy rail system. The
validation effort was first made by comparing model predictions against the NTD 2011
estimates, followed by a further discussion of modeling results. The NTD energy con-
sumption for the Chicago rail system was 36.44 kW h/V M (22.63 kW h/V · km) and
0.13 kW h/SM (0.08 kW h/S · km).

5.1. Energy consumption

A train on the Chicago ”L” system was tested by Georgia Tech researchers to collect
the trajectory data for energy prediction. The Chicago rail has the similar traction and

11
 100

90
Braking Regenerative Energy Efficiency (%)

80

70

60

50

40

30

20

10

0
−10 −9 −8 −7 −6 −5 −4 −3 −2 −1 0
Deceleration (m/s2)

Fig. 4: Regenerative braking efficiency varies as a function of the deceleration level

brake system, track infrastructure and vehicle aerodynamics compared to the Portland rail
system. The information for the testing vehicle is illustrated in Table 1. Compared to
the train running on the MAX Blue Line, the Chicago train has lower empty car weight,
seating capacity and the number of axles while more connecting cars. The testing route,
as demonstrated in Table B2, starts from the Kimball station and ends at the Merchandise
Mart station, covering 8.7 mi (14.01 km) section of the Chicago Brown Line. The vertical
layout of the section demonstrates a combination of upgrade (up to 2.37%) and downgrade
(up to -2.55%). The test was completed under the same weather and rail conditions as
those on the MAX Blue Line in order to ensure the identical bearing, track and weather
resistance between calibration and validation processes. The test results in the Chicago
Brown Line driving cycle with a maximum operational speed of 75 km/h and a total
duration of around 2000 s, as illustrated in Fig. 5. It is worth noting that the number of cars
and passenger load differ between the peak and the off-peak periods, resulting in different
train weights at different periods. Consequently, energy consumption was first estimated
for each period respectively, and then the average of the two periods was compared against
the NTD data.
Table 3 demonstrates that the model predictions are consistent with the NTD estimates,
generating a predicted error of 1.87% (kW h/V ·mi) and -2.31% (kW h/S ·mi). Not mod-
eling regenerative braking (”No regeneration” as shown in Table 3) results in a significant
prediction error (27.39% and 21.54%). Modeling the regenerative efficiency as a constant
also produces much higher prediction errors compared to the proposed model (16.11% and
10.77%). Furthermore, the constant assumption deems the model incapable of capturing
instantaneous energy regeneration so it is not suitable for a microscopic level analysis.

12
 80

70

60

50
Speed (km/h)

40

30

20

10

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (s)

Fig. 5: The driving cycle for the Chicago Brown line

Table 3: Validation on Energy Prediction

NTD Instantaneous regeneration Constant regeneration No regeneration

esti-
mates
Predicted Error Predicted Error Predicted Error
energy energy energy
kW h/V · 36.44 37.12 1.87% 42.31 16.11% 46.42 27.39%
mi
kW h/V · 22.63 23.06 26.28 28.83
km
kW h/S· 0.13 0.127 -2.31% 0.144 10.77% 0.158 21.54%
mi
kW h/S· 0.08 0.079 0.089 0.098
km

13
 Speed
80 Regeneration
No Regeneration
Speed (km/h)

60 Total Energy Consumption

40

20

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Instantaneous Energy (kW)

5000

−5000
200 400 600 800 1000 1200 1400 1600 1800 2000
5
x 10
Cumulative Energy (kW)

15

10

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (s)

Fig. 6: Chicago Brown Line: speed and electric power on the entire cycle

Fig. 6 demonstrates the adequacy of the model at an instantaneous level. The light
blue area represents the energy consumed for the entire cycle. The area delimited by
the blue edge line refers to the energy consumption without energy regeneration during
braking, and the red edge line represents the case considering energy regeneration. When
the train is in traction mode, the electric power is positive and the energy flows from the
overhead catenary to the wheels; alternatively, when the train is in brake mode, the power is
negative and the energy is sent back to the overhead catenary. The instantaneous variation
in the cumulative energy consumption demonstrates the ability of the model in adequately
predict electric consumption and energy recovery on an instantaneous basis. Fig. 6 also
reveals that energy recovery significantly reduces the overall power consumption.

5.2. Discussion of modeling results

The model predictions in Table 3 also demonstrate that regenerative braking achieves
an energy saving of 20% by reducing the energy consumption from 46.42 kW h/V · mi
(28.83 kW h/V · km) to 37.12 kW h/V · mi (23.06 kW h/V · km). This predicted energy
saving is consistent with what is reported in the literature [46–48], which demonstrated an
energy saving of up to 30%.
Furthermore, the proposed modeling approach is able to account for the impact of train
parameters (number of cars, seating capacity) on energy consumption. A comparative
analysis of predicted energy consumption between the MAX Blue Line and the Chicago
Brown Line demonstrates that the Chicago train consumes approximately 174% more en-
ergy in kW h/V · mi (13.57 vs. 37.12), while only consumes 20% more for per seating
mile (0.106 vs. 0.127). This demonstrates that the Chicago train has significantly more

14
 Table 4: Sensitivity of model predictions to railcar empty weight

Predicted energy consumption Energy consumption increment

Car weight Empty car
increment weight kW h/V · kW h/V · kW h/V · kW h/V ·
(ton) (ton) mi km mi km
-10 17.15 25.92 16.10 -11.20 -6.96
-5 22.15 31.52 19.58 -5.60 -3.48
0 27.15 37.12 23.06 0.00 0.00
5 32.15 42.72 26.53 5.60 3.48
10 37.15 48.32 30.01 11.20 6.96

per-train seats. The Chicago train has lower car seating capacity (49 as shown in Table 1)
yet more connecting cars, resulting in higher per-train seating.
In addition, the sensitivity of the model predictions to train weight and road grade was
analyzed. It should be noted that the energy consumption increments in Table 4 and Table 5
are relative to the base case in which the empty car weight is 27.15 tons and the number of
cars is 6 for the peak period and 4 for the off-peak period; and the increments in Table 6 are
relative to the case with level grade. As illustrated in Table 4, the original empty car weight
(27.15 tons) was either decreased or increased by 5 and 10 tons, respectively, generating
an identical energy consumption increase of 5.6 kW h/V ·mi (3.48 kW h/V ·km) for every
5 ton increment. For the original car weight, the number of cars for the base case was either
decreased or increased by 1 at a time. The increasing number of cars results in an identical
electric consumption increment of 6.19 kW h/V ·mi (3.84 kW h/V ·km), as demonstrated
in Table 5. Either increasing car weight or adding more connecting cars leads to linear
and identical growth of energy consumption. However, increasing the road grade, from
0% to 4% as illustrated in Table 6, produces a non-linear feature by having higher energy
consumption increments on steeper roads. For example, the energy consumption increases
by 9.2% (from 27.11 to 29.61 kW h/V · km) with the grade varying from 3% to 4%, while
increases by only 5.3% (from 22.25 to 23.44 kW h/V · km) with the grade varying from
0% to 1%. This implies that, on steeper roads, the identical gradient increase results
in larger energy consumption increases. The sensitivity analysis demonstrates the model’s
ability to capture energy consumption differences associated with vehicle weight and route
characteristics.
Finally, the MAX Blue Line driving cycle was applied to the Chicago train, generating
an average energy consumption of 23.10 kW h/V · mi (14.35 kW h/V · km), which is sig-
nificantly lower than that for the Chicago Brown Line cycle (37.12 kW h/V ·mi). A further
analysis, as illustrated in Fig. 7, demonstrates that the higher electric consumption for the

15
 Table 5: Sensitivity of model predictions to the number of railcars

Number of cars Predicted energy consumption Energy consumption increment

Number of cars
increment Peak Off- kW h/V · kW h/V · kW h/V · kW h/V ·
peak mi km mi km
-2 4 2 24.74 15.37 -12.38 -7.68
-1 5 3 30.93 19.21 -6.19 -3.84
0 6 4 37.12 23.06 0.00 0.00
1 7 5 43.31 26.90 6.19 3.84
2 8 6 49.50 30.75 12.38 7.68

Table 6: Sensitivity of model predictions to road grade

Road grade Predicted energy consumption Energy consumption increment

kW h/V · mi kW h/V · km kW h/V · mi kW h/V · km
0% 35.83 22.25 0.00 0.00
1% 37.74 23.44 1.91 1.19
2% 40.31 25.04 4.48 2.78
3% 43.65 27.11 7.82 4.86
4% 47.68 29.61 11.85 7.36

16
Chicago Brown Line Acceleration (m/s2) MAX Blue Line Acceleration (m/s2)
2

−1

−2

−3
0 1000 1000 2000 2000 3000 3000 4000 4000 5000 5000 6000 60007000

6
4
4
2
2

−2
−2
−4
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (s)

Fig. 7: Instantaneous acceleration levels

Chicago driving cycle is attributed to more aggressive running that produces higher accel-
eration levels. The model thus can reflect the energy consumption differences between rail
lines by accounting for the impact of operational conditions on energy prediction, which
is very important for multi-modal transit planning purposes.
Although the model was calibrated against the Portland light rail, it was demonstrated
to have good performance for the Chicago heavy rail as well, implying that similar lev-
els of model accuracy could be expected for other transit systems if there is no significant
difference in the energy consumption-related system components, such as propulsion tech-
nologies, brake systems, train aerodynamics and track infrastructure.

6. Conclusions
The paper develops an electric train energy consumption modeling framework consid-
ering instantaneous regenerative braking efficiency. The model is calibrated using data
from Portland, Oregon utilizing an unconstrained non-linear optimization procedure. The
validation effort is conducted using data from Chicago, Illinois by comparing model pre-
dictions against NTD estimates, followed by a further discussion on modeling results.
The results demonstrate that regenerative braking efficiency exponentially varies as
a function of the deceleration level, rather than an average constant as assumed in pre-
vious studies. The model predictions are consistent with the NTD estimates, producing
a predicted error of 1.87% and -2.31%. Not modeling regenerative braking results in a
significant prediction error (27.39% and 21.54%). It is also found that energy recovery re-
duces the overall power consumption by 20%, significantly improving the system energy
efficiency. The results also demonstrate that the proposed modeling approach is able to
capture the energy consumption differences associated with train, route and operational
characteristics, and thus is applicable for project-level analysis.

17
 The use of the proposed model does not require any internal engine data given that
the power exerted by a train is a function of the vehicle speed and acceleration which can
be collected using non-engine instrumentation (e.g. GPS). Accordingly, the model can be
easily implemented either in traffic simulation software, smartphone applications or eco-
transit programs. The model can thus support GHG emissions prediction and energy and
air quality analysis in rail transit.

7. Limitations and future research

The major limitation of this study is the limited availability of instantaneous trajectory
and energy consumption data. The authors only had access to trajectory data from the
Portland and Chicago transit agencies. Consequently, it is recommended that additional
data be gathered in the future to test the robustness of the model calibration procedure.
More importantly, field energy consumption data is needed to further verify the modeling
results.
Another limitation is that the model was only developed for trains from the Portland
and Chicago rail systems. Other transit systems may have different traction and brake
modes, track infrastructure and propulsion technologies which may lead to different cali-
bration results. Consequently, modeling other electric train modes is an interesting direc-
tion to be followed. Also, diesel-electric train modeling is another topic to be investigated
as an extension to this study.
Finally, a further study on model application is needed to develop rail transit simulation
models, smartphone eco-driving and eco-routing systems and eco-transit programs. These
applications will significantly support rail transit energy and air quality analysis.

8. Acknowledgements
The authors acknowledge the financial support provided by the TranLIV E University
Transportation Center and the Department of Energy ARPA-E Program. In addition, the
authors would like to acknowledge the help and data received from Dr. Randall Guensler
and his team at Georgia Tech.

9. References
[1] U.S. energy information administration (EIA). energy consumption by sector (2014).

[2] Davis SC , Diegel SW , Boundy RG . Transportation energy data book. Oak Ridge
National Laboratory, 2015.

18
 [3] EPA. Inventory of us greenhouse gas emissions and sinks: 1990-2013. Tech. rep..
EPA 430-R-15-004 (2015).

[4] Watanabe S , Koseki T . Energy-saving train scheduling diagram for automatically

operated electric railway. Journal of Rail Transport Planning & Management 5 (3)
(2015) 183–193.

[5] González-Gil A , Palacin R , Batty P . Sustainable urban rail systems: Strategies

and technologies for optimal management of regenerative braking energy. Energy
conversion and management 75 (2013) 374–388.

[6] Feng X . Optimization of target speeds of high-speed railway trains for traction en-
ergy saving and transport efficiency improvement. Energy Policy 39 (12) (2011)
7658–7665.

[7] Feng X , Mao B , Feng X , Feng J . Study on the maximum operation speeds of metro
trains for energy saving as well as transport efficiency improvement. Energy 36 (11)
(2011) 6577–6582.

[8] Fiori C , Ahn K , Rakha HA . Power-based electric vehicle energy consumption

model: Model development and validation. Applied Energy 168 (2016) 257–268.

[9] Gao DW , Mi C , Emadi A . Modeling and simulation of electric and hybrid vehicles.
Proceedings of the IEEE 95 (4) (2007) 729–745.

[10] How subway works (2007, http://science.howstuffworks.com/engineering/civil/subway3.htm).

[11] Hay WW . Railroad engineering. Vol. 1. John Wiley & Sons, 1982.

[12] Gbologah F , Xu Y , Rodgers M , Guensler R . Demonstrating a bottom-up frame-

work for evaluating energy and emissions performance of electric rail transit op-
tions. Transportation Research Record: Journal of the Transportation Research
Board (2428) (2014) 10–17.

[13] Puchalsky C . Comparison of emissions from light rail transit and bus rapid
transit. Transportation Research Record: Journal of the Transportation Research
Board (1927) (2005) 31–37.

[14] Messa C . Comparison of emissions from light rail transit, electric commuter rail, and
diesel multiple units. Transportation Research Record: Journal of the Transportation
Research Board (1955) (2006) 26–33.

19
[15] FTA, U.S. department of transportation. National Transit Annual
Database,http://www.ntdprogram.gov/ntdprogram/data.htm.

[16] Mittal RK . Energy intensity of intercity passenger rail. final report. Tech. rep..
Department of Transportation, Washington, DC (USA). Office of Univ. Research
(1977).

[17] Hopkins JB . Railroads and the environment estimation of fuel consumption in rail
transportation, volume 1: Analytical model. Tech. rep.. Federal Railroad Adminis-
tration. Final Report FRA-OR-D-75-74.I.Washington, DC (1975).

[18] Jong JC , Chang EF . Models for estimating energy consumption of electric trains.
Journal of the Eastern Asia Society for Transportation Studies 6 (2005) 278–291.

[19] Barrero R , Tackoen X , Van Mierlo J . Quasi-static simulation method for evaluation
of energy consumption in hybrid light rail vehicles. in: 2008 IEEE Vehicle Power
and Propulsion Conference. IEEE, 2008. pp. 1–7.

[20] Liu RR , Golovitcher IM . Energy-efficient operation of rail vehicles. Transportation

Research Part A: Policy and Practice 37 (10) (2003) 917–932.

[21] Lombardi EJ . Engineering tests performed on the x2000 and ice high speed train-
sets. in: Railroad Conference, 1994., Proceedings of the 1994 ASME/IEEE Joint (in
Conjunction with Area 1994 Annual Technical Conference). IEEE, 1994. pp. 13–21.

[22] Álvarez A . Energy consumption and emissions of high-speed trains. Transportation

Research Record: Journal of the Transportation Research Board (2159) (2010) 27–
35.

[23] Kim K , Chien SIJ . Optimal train operation for minimum energy consumption con-
sidering track alignment, speed limit, and schedule adherence. Journal of Transporta-
tion Engineering 137 (9) (2010) 665–674.

[24] Sicre C , Cucala AP , Fernández A , Lukaszewicz P . Modeling and optimizing

energy-efficient manual driving on high-speed lines. IEEJ Transactions on Electrical
and Electronic Engineering 7 (6) (2012) 633–640.

[25] Su S , Li X , Tang T , Gao Z . A subway train timetable optimization approach based

on energy-efficient operation strategy. IEEE Transactions on Intelligent Transporta-
tion Systems 14 (2) (2013) 883–893.

20
[26] González-Gil A , Palacin R , Batty P , Powell J . A systems approach to reduce urban
rail energy consumption. Energy Conversion and Management 80 (2014) 509–524.

[27] Li X , Lo HK . An energy-efficient scheduling and speed control approach for metro

rail operations. Transportation Research Part B: Methodological 64 (2014) 73–89.

[28] Golovitcher IM . Energy efficient control of rail vehicles. in: Systems, Man, and
Cybernetics, 2001 IEEE International Conference on. Vol. 1. IEEE, 2001. pp. 658–
663.

[29] Jenks CW , Goldstein LD , Avery AP , Delaney EP , Lamberton S . National coopera-

tive rail research program (ncrrp) report 3: Comparison of passenger rail energy con-
sumption with competing modes. Tech. rep.. Transportation Research Board, Wash-
ington DC (2015).

[30] Argonne national laboratory. autonomie (2015).

[31] Lewis AM , Kelly JC , Keoleian GA . Evaluating the life cycle greenhouse gas emis-
sions from a lightweight plug-in hybrid electric vehicle in a regional context. in: 2012
IEEE International Symposium on Sustainable Systems and Technology (ISSST).
IEEE, 2012. pp. 1–6.

[32] Lewis AM , Kelly JC , Keoleian GA . Vehicle lightweighting vs. electrification: Life

cycle energy and ghg emissions results for diverse powertrain vehicles. Applied En-
ergy 126 (2014) 13–20.

[33] De Gennaro M , Paffumi E , Scholz H , Martini G . Gis-driven analysis of e-mobility

in urban areas: an evaluation of the impact on the electric energy grid. Applied En-
ergy 124 (2014) 94–116.

[34] Rambaldi L , Bocci E , Orecchini F . Preliminary experimental evaluation of a four

wheel motors, batteries plus ultracapacitors and series hybrid powertrain. Applied
Energy 88 (2) (2011) 442–448.

[35] Abousleiman R , Rawashdeh O . Energy consumption model of an electric vehicle.

in: Transportation Electrification Conference and Expo (ITEC), 2015 IEEE. IEEE,
2015. pp. 1–5.

[36] Rajashekara K , Rathore AK . Power conversion and control for fuel cell systems
in transportation and stationary power generation. Electric Power Components and
Systems 43 (12) (2015) 1376–1387.

21
[37] Wang J , Rakha HA . Modeling fuel consumption of hybrid electric buses: Model
development and comparison with conventional buses. Transportation Research
Record: Journal of the Transportation Research Board (2539) (2016) 94–102.
[38] Association ARE , et al.. Manual for Railway Engineering 1980-1981. American
Railway Engineering Association, 1980-1981.
[39] Schmidt EC . Freight Train Resistance: Its Relation to Car Weight. no. 43. The Uni-
versity, 1934.
[40] Tuthill JK . High Speed Freight Train Resistance: Its Relations to Average Car
Weight. no. 32. The University, 1948.
[41] Gould G , Niemeier D . Review of regional locomotive emission modeling and the
constraints posed by activity data. Transportation Research Record: Journal of the
Transportation Research Board (2117) (2009) 24–32.
[42] Volt C . Chevy volt tech watch: regenerative braking (2011).
[43] Pham KD , Thomas RS , Ramirez X . Traction power supply for the portland in-
terstate max light rail extension. in: Transportation Research Circular E-C058: 9 th
National Light Rail Transit Conference. 2004. pp. 669–677.
[44] Ceraolo M , Lutzemberger G . Stationary and on-board storage systems to enhance
energy and cost efficiency of tramways. Journal of Power Sources 264 (2014) 128–
139.
[45] Frilli A , Meli E , Nocciolini D , Pugi L , Rindi A . Energetic optimization of regen-
erative braking for high speed railway systems. Energy Conversion and Management
129 (2016) 200–215.
[46] International union of railways. regenerative braking for 50 hz, 25 kv systems (2002,
http://www.railway-energy.org/tfee/index.php).
[47] Sudhir, M.D., braking green for mumbai trains (2012,
http://blog.siemens.co.in/?p=4559).
[48] Bombardier. Bombardier environmental technologies help delhi metro
achieve world-first under united nations-backed climate change initia-
tive (http://csr.bombardier.com/en/media-center/ csr-news/25-bombardiers-
environmental-technologies-help-delhimetro-achieve-world-first-under-united-
nations-backed climate changeinitiative. Accessed July 26, 2013.).

22
Appendix A. Model Parameter Annotation

23
 Table A1: Description of model parameters

Name Description Unit

E Annual energy consumption kW h
Ep Annual energy consumed per passenger kilometer kW h/P · km
Es Annual energy consumed per seating kilometer kW h/S · km
Ev Annual energy consumed per vehicle kilometer kW h/V · km
Mp Annual passenger kilometer P · km
Ms Annual seating kilometer S · km
Mv Annual vehicle kilometer V · km
C Train capacity -
β Line loss factor -
EC(t) Instantaneous energy consumption kW
ECre (t) Instantaneous energy regeneration kW
P (t) Instantaneous tractive power kW
α01 Head-end power (HEP) kW
α02 Fraction of HEP (0.05 is suggested by [12]) kW
β01 , β02 Dummy variable -
ηre Instantaneous regenerative efficiency -
ECd Trip energy consumption per unit distance kW h/V · km
F (t) Instantaneous tractive force N
u(t) Instantaneous speed km/h
wp Weight per railcar axle ton
np Number of axles per railcar -
K Train drag coefficient (0.07 suggested by [11]) -
θ Road grade %
L Distance a train moved in one second m
M Total train weight ton
α Regenerative efficiency model parameter -
a(t) Instantaneous deceleration level m/s2
d distance covered by the entire driving cycle km
ECN T D NTD energy rate kW h/V · km

24
Appendix B. Track information

Table B1: Track information for the MAX Blue Line

Milepost Elevation
Station Name mile kilometer f oot meter Grade to Next Station
Hatfield Government Center Station 0.000 0.000 200.000 60.960 0.00049
Hillsboro Central /SE 3rd TC 0.360 0.579 200.940 61.247 -0.00122
Tuality Hospital/SE 8th 0.701 1.128 198.745 60.577 -0.00580
Washington/SE 12th 1.117 1.799 185.995 56.691 0.00320
Fair Complex/Hillsboro Airport 2.348 3.781 206.789 63.029 -0.00079
Hawthorn Farm 3.116 5.016 203.609 62.060 0.00310
Orenco/NW 231st 3.845 6.190 215.559 65.702 -0.00095
Quatama/NW 205th 5.265 8.477 208.439 63.532 0.00095
Willow Creek/SW 185th 6.250 10.063 213.379 65.038 0.00047
Elmonica/SW 170th 7.273 11.709 215.894 65.804 0.00210
Merlo/SW 158th 7.860 12.654 222.419 67.793 0.00037
Beaverton Creek 8.409 13.539 223.499 68.122 -0.00532
Millikan Way 9.110 14.667 203.829 62.127 0.00122
Beaverton Central 9.848 15.856 208.594 63.579 -0.00124
Beaverton TC 10.170 16.374 206.494 62.939 0.02039
Sunset TC 12.197 19.637 424.711 129.452 0.00325
Washington Park 15.426 24.836 480.081 146.329 -0.04455
Goose Hollow/SW Jefferson 16.818 27.077 152.648 46.527 -0.01191
Kings Hill/SW Salmon 17.045 27.443 138.358 42.171 0.00804
JELD-WEN Field 17.159 27.626 143.183 43.642 0.00690
13 th st 17.424 28.053 152.838 46.585 -0.02200
Galleria/SW 10th 17.557 28.266 137.438 41.891 -0.02200
Pioneer Square North 17.727 28.541 117.638 35.856 -0.02200
Mall/SW 5th 17.841 28.724 104.438 31.833 -0.01100
SW 3rd 18.030 29.029 93.438 28.480 -0.00895
Oak/SW 1st 18.220 29.334 84.488 25.752 -0.00117
Skidmore Fountain 18.447 29.700 83.088 25.325 0.00334
Old Town/Chinatown 18.574 29.904 85.328 26.008 0.03266
Rose Quarter TC 19.085 30.727 173.498 52.882 0.04261
Convention Center 19.218 30.941 203.328 61.974 0.01174
NE 7th 19.545 31.468 223.638 68.165 0.00500
Lloyd Center/NE 11th 19.634 31.611 225.988 68.881 0.00051

25
Hollywood/NE 42nd TC 21.377 34.417 230.673 70.309 0.00681
NE 60th 22.381 36.033 266.773 81.312 0.00398
NE 82nd 23.782 38.289 296.213 90.286 0.02047
Gateway/NE 99th TC 24.559 39.540 380.133 115.864 -0.01174
E 102nd 25.402 40.896 327.873 99.936 0.00250
E 122nd 26.585 42.802 343.468 104.689 0.00331
E 148th 27.708 44.610 363.068 110.663 -0.00660
E 162nd 28.371 45.678 339.968 103.622 -0.00808
E 172nd 28.864 46.470 318.968 97.221 -0.00552
E 181st 29.299 47.172 306.268 93.350 0.00275
Rockwood/E 188th 29.678 47.782 311.768 95.027 0.00584
Ruby Junction/E 197th 30.227 48.666 328.718 100.193 0.00866
Gresham City Hall 31.610 50.892 391.918 119.456 0.01529
Gresham Central TC 32.121 51.715 433.198 132.039 0.01419
Cleveland 32.538 52.386 464.408 141.551 0.01419

26
 Table B2: Track information for the Chicago Brown Line

Milepost Elevation
Station Name mile kilometer f oot meter Grade to Next Station
Kimball 0.000 0.000 548.314 167.126 -0.0255
Kedzie 0.252 0.406 514.337 156.770 -0.0169
Francisco 0.626 1.008 481.030 146.618 -0.0011
Rockwell 0.983 1.583 479.016 146.004 0.0094
Western 1.322 2.128 495.863 151.139 0.0025
Damen 1.766 2.844 501.795 152.947 0.0106
Montrose 2.245 3.614 528.681 161.142 0.0196
Irving Park 2.747 4.423 580.702 176.998 -0.0223
Addison 3.222 5.187 524.777 159.952 0.0072
Paulina 3.592 5.784 538.858 164.244 0.0067
Southport 3.955 6.368 551.759 168.176 -0.0030
Belmont 4.692 7.554 539.964 164.581 0.0125
Wellington 4.897 7.885 553.511 168.710 -0.0052
Diversey 5.155 8.300 546.453 166.559 -0.0187
Fullerton 5.642 9.084 498.304 151.883 -0.0071
Armitage 6.106 9.831 480.804 146.549 0.0237
Sedgwick 7.180 11.560 615.272 187.535 -0.0110
Chicago 8.230 13.250 554.058 168.877 0.0091
Merchandise Mart 8.711 14.024 577.077 175.893 0.0091

27