Sunteți pe pagina 1din 9

Applied Energy 170 (2016) 394402

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Fuel consumption model for conventional diesel buses


Jinghui Wang, Hesham A. Rakha
Charles E. Via, Jr. Department of Civil and Environmental Engineering, Virginia Tech, Blacksburg, VA 24061, USA

h i g h l i g h t s
 The research collects bus fuel consumption data for diesel buses.
 Models are developed to compute the fuel consumption levels of buses.
 The optimum bus fuel economy cruising speeds range between 40 and 50 km/h.
 The model is more consistent with empirical observations compared to the MOVES and CMEM models.

a r t i c l e

i n f o

Article history:
Received 26 October 2015
Received in revised form 16 February 2016
Accepted 22 February 2016
Available online 12 March 2016
Keywords:
Heavy duty diesel
Transit bus
Fuel consumption modeling
VT-CPFM

a b s t r a c t
Existing bus fuel consumption models produce a bangbang type of control, implying that drivers
would have to either accelerate at full throttle or brake at full braking in order to minimize their fuel
consumption levels. This is obviously not correct. The paper is intended to enhance bus fuel consumption
modeling by circumventing the bangbang control problem using the Virginia Tech Comprehensive
Power-based Fuel consumption Model (VT-CPFM) framework. The model is calibrated for a series of
diesel-powered buses using in-field second-by-second data because of a lack of publicly available bus fuel
economy data. The results reveal that the bus fuel consumption rate is concave as a function of vehicle
power instead of convex, as was the case with light duty vehicles. The model is calibrated for an entire
bus series and demonstrated to accurately capture the fuel consumption behavior of each individual
bus within its series. Furthermore, the model estimates are demonstrated to be consistent with
in-field measurements. The optimum fuel economy cruising speeds range between 40 and 50 km/h,
which is slightly lower than that for gasoline-powered light duty vehicles (6080 km/h). Finally,
the model is demonstrated to capture transient fuel consumption behavior better than the Motor
Vehicle Emission Simulator (MOVES) and produces a better fit to field measurements compared to the
Comprehensive Modal Emission Model (CMEM).
2016 Elsevier Ltd. All rights reserved.

1. Introduction
The U.S. Bureau of Transportation Statistics (US BTS) reported
that bus fuel consumption had been continually increasing from
827 million gallons/year to 2059 million gallons/year between
1960 and 2012. As demonstrated by Greene et al. [1], transit buses
produced the highest level of energy consumption compared to
other traffic modes such as passenger cars, light-duty trucks, air
transportation, rail transit. Consequently, improving bus fuel
efficiency is of value to reduce energy consumption and CO2
production from the transportation perspective.
An accurate and efficient fuel consumption model is needed to
capture fuel consumption behavior and quantify potential
improvements in fuel efficiency induced by implementing
Corresponding author.
E-mail address: hrakha@vt.edu (H.A. Rakha).
http://dx.doi.org/10.1016/j.apenergy.2016.02.124
0306-2619/ 2016 Elsevier Ltd. All rights reserved.

eco-friendly strategies, such as developing eco-routing [24] or


eco-driving systems [57] and the use of alternative fuel techniques [812]. Numerous models have been developed from
macroscopic to mesoscopic to microscopic levels, such as MOBILE
[13], Emission Factors (EMFAC) [14], the VT-Meso model [15],
(CMEM) [16], Motor Vehicle Emissions Simulator (MOVES) [17],
vehicle transient emissions simulation software (VeTESS) [18],
NetSim [19], VERSIT [20], the passenger car and heavy duty
emissions model (PHEM) [21], EMIssions from Traffic (EMIT)
[22], the VT-Micro model [23], the Virginia Tech Comprehensive
Power-based Fuel consumption Model (VT-CPFM) [24], and the
Virginia Tech Comprehensive Power-based Energy consumption
Model (VT-CPEM) [25]. Macro-models, however, may produce
unreliable estimates given that they are incapable of accounting
for transient vehicle behavior along a route [26]. Instead, micromodels are better suited for capturing fuel consumption effects
associated with a wide range of vehicle dynamics.

395

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402

Despite the models described above, only a few are applicable


to transit buses. For example, the CMEM was originally targeted
for light duty vehicles (LDVs), but has been expanded to
medium- and heavy-duty vehicles [27]. This model is a physically
power-based model that separates the entire modeling process
into different components which consist of a power demand
model, an engine speed estimator, a fuel rate model, an engineout emission component and an after-treatment component. The
first three components are designed to generate fuel consumption
rates. The bus fuel rate is estimated using Eq. (1) [27]:



i
K  N  V gP h
 1 b1  N  N0 2
FR
43:2

where FR is the fuel rate in g/s, P is engine power in kW, K is engine


friction factor, N is the engine speed (revolutions per second), V is
the engine displacement (l), g is a measure of the engine efficiency
for diesel engines which equals to 0.45, b1 and C are model coefficients and 43.2 kJ/g is the lower heating value of a typical diesel
fuel. Using Eq. (1), CMEM basically uses a linear relationship
between fuel consumption and vehicle power, which may produce
a bangbang type of control system (the bangbang control system
will be mathematically introduced in Section 2). A bangbang control may arise when the partial derivative of fuel consumption with
respect to the vehicle power is not a function of vehicle power. This
results in the optimum fuel economy control being achieved at full
throttle acceleration or at full braking, whereby drivers drive as
aggressively as possible to minimize their fuel consumption levels,
which is obviously not correct in reality. The PHEM model was first
developed for heavy duty vehicles (HDVs) and thereafter extended
to passenger cars. This model also provides a bangbang control
system since the fuel consumption is affine to vehicle power. North
[28] reported that PHEM provided less accurate predictions of emissions than the INTEGRATION model which uses the VT-Micro model
as an estimator. Although the VT-Micro model circumvents the
bangbang control, it requires a large amount of field data for calibration purposes, which is not a cost-effective and time-efficient
approach. Another attempt at bus fuel consumption modeling is
MOVES, complimented by the Physical Emission Rate Estimator
(PERE) [29], which is designed by EPA to fill data gaps for MOVES.
PERE is basically a power-based modeling approach by which fuel
rates are assumed to be determined by vehicle specific power
(VSP), engine speed, and vehicle mass. Although it can provide
robust estimates, the use of MOVES is time-consuming since users
have to construct numerous input profiles through a graphical user
interface every time they use it. Other microscopic models, such as
VeTESS and VERSIT, either provide poor predictions (VeTESS estimates generally have an accuracy between 10% and 20% [30]) or
produce a bangbang control (VERSIT). EMIT is not applicable when
the road grade is not equal to zero [28,31], ignoring the impact of
road grade on fuel consumption.
Overall, the existing models either produce a bangbang type of
control system or cannot be easily calibrated or efficiently used.
Consequently, a simple, accurate and efficient model is needed.
The Virginia Tech Comprehensive Power-based Fuel consumption
Model (VT-CPFM) framework was first developed by Rakha et al.
[24] and successfully circumvented bangbang control in the fuel
consumption modeling practice by characterizing fuel consumption as a second-order polynomial function of the vehicle power.
Furthermore, the model offers a unique ability to be calibrated
using publicly available data, which makes it more efficient and
cost-effective in calibration compared to VT-Micro model. However, this model was originally targeted at light duty vehicles
(LDVs). Consequently, the objective of this study is to expand
the VT-CPFM framework to transit buses in order to eliminate

bangbang control in the family of existing bus fuel consumption


models.
2. A bangbang control system
Minimizing the fuel consumption level is an optimal control
problem that attempts to compute the optimal solution with the
control variable restricted to being between a lower and an upper
bound. In optimal control problems, a bangbang solution may
occur when a control switches abruptly from one extreme to the
other. To mathematically give a complete picture of the bangbang
control, the minimum-fuel problem is described in Eq. (2), all of
which are derived from Pontryagins Maximum Principle (one of
the most prevailing approaches to address optimal control
problems) [32,33]:

te

min
P;t e

m_ f Ptdt

2a

subject to : Pmin t 6 Pt 6 Pmax t

2b

where m_ f is the fuel mass flow rate [kg/s], P is the vehicle power
[kW], Pmin and Pmax are the minimum and maximum of vehicle
power respectively [kW].
Vehicle power (P) is the control variable in the problem above.
The optimal control is the control variable that minimizes m_ f .
Namely, the optimal power is obtained where the fuel consumption rate is at the minimum level. If m_ f is a linear function of P,
the optimal solution would be achieved at either P min or Pmax . This
demonstrates that an acceleration of a vehicle should be at full
throttle or full braking to achieve the minimum fuel consumption level, which is not realistic. Consequently, a model with higher
order is needed to circumvent the linear relationship and thus to
avoid bangbang control.
3. The VT-CPFM framework
The VT-CPFM was originally developed for LDVs. A secondorder polynomial function was used as the modeling framework
since the relationship between fuel consumption rates and positive
power conditions exhibited a parabolic characteristic [24]. The
model was subsequently calibrated and validated by Park et al.
[34] using data collected using an on-board diagnostic (OBD)
reader, demonstrating that the model estimates had a good agreement with field measurements. This model, however, has not been
expanded to diesel buses. The data in Fig. 1, gathered using a HEM
data logger (the details of HEM will be introduced in Section 4),
indicate that for positive power conditions the fuel consumption
exhibits an approximation of a parabolic relationship to the vehicle
power and for negative conditions it remains nearly constant. This
yields that bus fuel consumption is related in a similar way to the
vehicle power as LDVs. Consequently, the VT-CPFM is applicable to
model diesel bus fuel consumption levels, whereas the accuracy of
the model remains to be studied in the following sections.
3.1. VT-CPFM modeling structure
The formulas of the two VT-CPFM models (VT-CPFM-1 and
VT-CPFM-2) are presented in Eqs. (3) and (4) respectively.

FCt
(
FCt

a0 a1 Pt a2 Pt2 ; 8Pt P 0
a0 ;
8Pt < 0
b0 xt b1 Pt b2 Pt2 ; 8Pt P 0
b0 xidle ;

8Pt < 0

396

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402

and 43,200,000 J/kg for diesel fuel); N is the number of engine


cylinders; F city and F hwy (l) are the fuel consumed for the EPA city
and highway drive cycles (computed as Eqs. (10) and (11)); P city

0.012

and P2city are the sum of the power and power squared over the

Fuel Consumption Rate(l/s)

0.01

EPA city cycle (computed as Eqs. (12) and (13)), P hwy and P2hwy are
calculated in the similar manner; T city and T hwy are the duration
of EPA city and highway drive cycles (s); FEcity and FEhwy are the fuel
economy estimates for the EPA city and highway cycles (km/l).

0.008

a0

0.006

P mfo xidle d
22164QN


a2

0.004

 

P
P
F city  F hwy P city  T city  T hwy P city a0
hwy

hwy

P2city  P2hwy P city

hwy

0.002

0
400

a1

300

200

100

100

200

300

F hwy  T hwy a0  P2hwy a2


Phwy

F city

3:7854  17:663 41:5546

1:6093  FEcity
FEcity

10

F hwy

3:7854  16:4107 38:6013

1:6093  FEhwy
FEhwy

11

Vehicle Power(kW)
Fig. 1. Vehicle power vs. fuel consumption functional form.

where FCt is the instantaneous fuel consumption rate [l/s];

a0 ; a1 ; a2 , and b0 ; b1 ; b2 are vehicle-specific model coefficients


that should be calibrated for each vehicle; x is the engine rotational
speed and xidle is the engine idling speed. VT-CPFM-2 requires
engine data (e.g. gear ratio.) for calibration in addition to external
data (e.g. speed, acceleration.). Due to the lack of engine data, only
VT-CPFM-1 is used to model bus fuel consumption in this study. In
the case of the VT-CPFM-1, the vehicle power is computed as [35]:

Rt 1 k 0:0025nv t mat
3600gd
2

Pt

v t

where Pt is the vehicle power (kW); Rt is the vehicle resistance


force (N); k is the mass factor accounting for rotational masses, a
value of 0.1 is used for heavy duty vehicles (HDVs) [36]; n is the
term related to gear ratio, which is assumed to be zero due to the
lack of gear data; m is the total weight of buses (kg) (including
bus curb weight and passenger weight); at is the instantaneous
acceleration (m/s2) and v t is the vehicle speed in the unit of
km/h; gd is the driveline efficiency. Rt is computed as the sum
of aerodynamic, rolling and grade resistance forces as expressed
in Eq. (6) [37], where q is the air density at sea level at a temperature of 15 C (59 F) (equal to 1.2256 kg/m3), C D is the vehicle drag
coefficient (unitless), the correction factor for altitude (unitless) is
calculated by 1  0:085H (H is the altitude in the unit of km), Af is
the frontal area of buses (m2 ), C r ; c1 and c2 are rolling resistance
parameters (unitless), Gt is the road grade.

Rt

q
Cr
C D C h Af v t2 9:8066m
c1 v t c2
25:92
1000
9:8066mGt

3.2. Discussion of model calibration


As aforementioned, the VT-CPFM model can be calibrated using
publicly available data without gathering numerous on-road or inlab measurements. For LDVs, the vehicle-specific coefficients a0 ; a1 ,
and a2 can be estimated using Eqs. (7)(9), where Pmfo is the idling
fuel mean pressure (400,000 Pa); d is the engine displacement (l);
Q is the fuel lower heating value (43,000,000 J/kg for gasoline fuel

Pcity

T hwy
T city
X
X
Pt and Phwy
Pt
t0

P2city

T hwy
T city
X
X
Pt2 and P2hwy
Pt2
t0

12

t0

13

t0

Nonetheless, transit buses do not report fuel economy for standard drive cycles (e.g. the EPA highway and city drive cycles),
implying that the FEcity and FEhwy are not available. This limits the
use of publicly available data to model diesel bus fuel consumption
levels. Regression analysis is thus considered as an alternative
technique to calibrate the model using on-road data. Although
the publicly available data have not been used in this paper, the
introduction of Eqs. (7)(13) gives some insights into the possible
calibration effort that could be applied if the vehicle economy is
reported in the future.
4. Model calibration
A regression-based approach is proposed to calibrate the VTCPFM model for diesel buses. Mass field data were collected by test
driving the buses around the town of Blacksburg, VA. The test was
conducted on two road sections: US 460 business (highway with a
speed limit of 65 mi/h (104 km/h)) and local streets (with the
speed limit from 25 mi/h to 45 mph (4072 km/h)) in order to
cover a wide range of real-world driving conditions. Specifically,
the test route comprised a variety of uphill and downhill sections,
and thus provided a suitable environment to test different engine
load conditions. The collected data were separated into two subsets for each bus. The first set which comprised 6070% percent
of the data was set for calibration, and the remaining data were
used for validation purposes.
A total of 14 conventional diesel buses were tested, as summarized in Table 1. The tested buses were classified into four series,
specified as 19xx, 62xx, 630x, and 632x, based on the vehicle
model year, engine brand and model, transmission, horsepower
and curb weight. It should be noted that all tested buses are
equipped with automatic transmission with 6 gears. Within the
same series, buses have the identical vehicle properties. Edwardes

397

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402


Table 2
Parameters required for model calibration.

Table 1
Bus specific information.
Bus
no.

Year

Engine
brand and
model

Transmission
make and model

Horsepower

1911

2009

Cummins
ISL-07

250330

28,300

1912

2009

Cummins
ISL-07

250330

28,300

1913

2009

Cummins
ISL-07

Allison
Transmission
B400R Gen 4
Allison
Transmission
B400R Gen 4
Allison
Transmission
B400R Gen 4
Allison
Transmission
B400R Gen 4
Allison
Transmission
B400R Gen 4
Allison
Transmission
B400R Gen 4
Allison
Transmission
B400R Gen 4
Allison
Transmission
B400 Gen 4
Allison
Transmission
B400 Gen 4
Allison
Transmission
B400 Gen 4
Allison
Transmission
B400 Gen 4
Allison
Transmission
B400 Gen 4
Allison
Transmission
B400 Gen 4
Allison
Transmission
B400 Gen 4

250330

28,300

1919

2009

Cummins
ISL-07

1920

2009

Cummins
ISL-07

1921

2009

Cummins
ISL-07

1923

2009

Cummins
ISL-07

6201

2012

Cummins
ISL-2010

6203

2012

Cummins
ISL-2010

6204

2012

Cummins
ISL-2010

6307

2013

Cummins
ISL-2010

6308

2013

Cummins
ISL-2010

6323

2013

Cummins
ISL-2010

6324

2013

Cummins
ISL-2010

Empty
weight
(lb)

250330

28,300

250330

28,300

250330

28,300

250330

28,300

280330

26,750

280330

26,750

280330

26,750

280330

26,750

280330

26,750

280330

39,675

280330

39,675

Parameter

Value

Source

Drag coeff (C D )
Altitude correction factor (C h )
Bus frontal area (Af )
Vehicle speed (v)
Ridership
Mass (m)
Rolling coefficient (C r )
c1
c2
Road grade (G)
Acceleration (a)
Driveline efficiency (gd )

0.8
/
6.824 m2
/
/
/
1.25
0.0328
4.575
/
/
0.95

Blacksburg Transit
Computed by field
Blacksburg Transit
Field data
Field data
Computed by field
[37]
[37]
[37]
Computed by field
Computed by field
[37]

data

data
data

Table 3
Series- and individual-models.
Bus no.

a0

a1

a2

19xx series
62xx series
630x series
632x series
1911
1912
1913
1919
1920
1921
1923
6201
6203
6204
6307
6308
6323
6324

1.606e03
1.076e03
9.356e04
1.211e03
1.546e03
1.499e03
1.307e03
1.599e03
1.728e03
1.323e03
1.931e03
1.090e03
8.516e04
1.083e03
9.109e04
1.132e03
1.230e03
1.192e03

1.067e04
7.294e05
8.323e05
1.092e04
1.139e04
1.079e04
1.104e04
1.288e04
1.128e04
1.095e04
1.195e04
7.585e05
7.262e05
7.881e05
8.319e05
8.574e05
1.125e04
1.059e04

2.849e07
2.149e07
2.694e07
2.946e07
4.498e07
2.665e07
3.749e07
4.579e07
3.206e07
2.251e07
3.198e07
3.355e07
2.244e07
2.110e07
2.489e07
4.858e07
4.154e07
1.738e07

0.012
19XX series
62XX series

Fuel Consumption Rate(l/s)

0.01

[38] demonstrated that the magnitude of the model coefficients


significantly differed between bus series, implying that the model
property mainly depended on bus-specific parameters. Namely,
the model may exhibit similar features within the same bus classification. Consequently, calibration was originally conducted for
each entire bus series (series model). Furthermore, the model
was also calibrated for each individual bus (individual model) in
order to capture the applicability of series models to each individual bus by comparing the performance of the two types of models.
For data collection, the Hydraulics + Electrical + Mechanical
(HEM) data logger, manufactured by IOSiX LLC and sold by HEM
Data Corporation, was used since it is small in size and capable
of collecting data autonomously without any maintenance. The
data logger was plugged into the J1939 port and installed between
the equipment cabinet behind the driver and the window. The data
were collected from ignition-on to ignition-off and saved on a
micro SD card, and then uploaded to a server via WiFi automatically (for more details of HEM, see [39]). In this study, up to 46
variables were collected while test driving the buses, six of which
were extracted for calibration purposes, namely: the time stamp,
vehicle speed, fuel consumption rate, latitude, longitude, and altitude. The data were recorded at a frequency of 2 Hz or 5 Hz. In
order to generate the second-by-second record, the original data
were combined by averaging the values of the data points within

Authority
data
Authority

630X series
632X series

0.008

0.006

0.004

0.002

50

100

150

200

Vehicle Power(kW)
Fig. 2. Illustration of fuel consumption models for each bus series.

each individual second. The null data, which had a value of 0 for
longitude/latitude/altitude, were removed from the data.
Table 2 summarizes the input parameters for model calibration.
Some of the parameters, such as rolling resistance coefficients and
driveline efficiency, can be obtained either from a transit agency or

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402

Fuel Consumption Rate(l/s)

398

0.014

Measurements
Estimated

0.012
0.01
0.008
0.006
0.004
0.002
0

200

400

800

600

1000

1200

1400

1600

Time(s)

Fuel Consumption Rate(l/s)

(a) 19XX Series


3

x 10

Measurements
Estimated

7
6
5
4
3
2
1
0

200

400

600

800

1000

1200

1400

1600

Time(s)

Fuel Consumption Rate(l/s)

(b) 62XX Series


0.01

Measurements
Estimated

0.008
0.006
0.004
0.002
0

200

400

600

800

1000

1200

1400

1600

Time(s)

Fuel Consumption Rate(l/s)

(c) 630X Series


0.012

Measurements
Estimated

0.01
0.008
0.006
0.004
0.002
0

200

400

800

600

1000

1200

1400

1600

Time(s)

(d) 632X Series


Fig. 3. Model validation at the instantaneous fuel consumption level.

from the literature; while some of the parameters should be computed based on field data. The total mass (m) is the sum of the bus
curb weight and passenger weights which are computed as the dot
product of the ridership and the weight of individual passengers. It
should be noted that 170 lb (76.5 kg) was assumed to be the average passenger weight in this study due to a lack of field data. Road
grade was computed using Eq. (14):

Altt Dt  Altt
Gt q
2
Dt Dt  Dt2  Altt Dt  Altt

14

with Alt the altitude at t and t Dt and D the distance a bus


travels in one second. Given that the measured altitude was not
accurate, GIS tool was used to correct the original data.

399

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402


Table 4
R-squared values for validation.

Table 5
Optimum fuel economy cruise speed.

Bus no.

Series model validation (R2 )

Individual model validation (R2 )

Bus no.

Optimum cruise speed (km/h)

1911
1912
1913
1919
1920
1921
1923
6201
6203
6204
6307
6308
6323
6324

0.820
0.813
0.753
0.747
0.801
0.856
0.767
0.712
0.839
0.798
0.756
0.724
0.801
0.796

0.820
0.812
0.726
0.747
0.801
0.853
0.767
0.788
0.840
0.798
0.755
0.723
0.792
0.796

19xx series
62xx series
630x series
632x series
1911
1912
1913
1919
1920
1921
1923
6201
6203
6204
6307
6308
6323
6324

46
46
42
42
46
45
43
44
47
42
48
48
42
45
41
45
42
40

Vehicle power was computed using Eq. (5). The relationship


between the estimated power and the measured fuel consumption
rate (as shown in Fig. 1) indicates that many outliers remained to
be removed. Three prevailing statistical techniques were chosen
for outlier elimination: studentized residual, leverage and cooks
distance. The studentized residual is an extremely good way in
measuring the level of an outlier in the vertical direction (ydirection) a point is; while leverage is used to detect outliers of
the independent variables (x-direction). The cooks distance, as a
combination of studentized residual and leverage, is a commonly

19XX series

62XX series

Fuel Consumption(l/km)

Fuel Consumption(l/km)

used measurement to detect the data points resulting in strong


influence on regression modeling.
Table 3 illustrates the developed series- and individual-models.
Interestingly, the second-order parameters a2 are negative, suggesting that the shape of the bus fuel consumption relationship
is concave as a function of the vehicle power unlike LDVs, which
exhibit a convex relationship. Furthermore, the order of magnitude

5
4
3
2

3.5
3
2.5
2
1.5
1

1
0.5
0

10

20

30

40

50

60

70

80

90

100

Cruise Speed(km/h)

10

20

30

50

60

70

80

90

100

80

90

100

Cruise Speed(km/h)
632X series

630X series
3.5

4.5
4

Fuel Consumption(l/km)

Fuel Consumption(l/km)

40

2.5
2
1.5
1

3.5
3
2.5
2
1.5
1

0.5
0

0.5
0

10

20

30

40

50

60

70

80

90

100

10

20

30

Cruise Speed(km/h)
Fig. 4. Impact of cruise speed on fuel consumption rates.

40

50

60

70

Cruise Speed(km/h)

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402

Fuel Consumption Rate (l/s)

400

0.012

Measurements
VTCPFM

0.01

CMEM
MOVES

0.008
0.006
0.004
0.002
0

1000

500

1500

Time (s)

Fuel Consumption Rate (l/s)

0.014
Measurements
VTCPFM

0.012

CMEM
MOVES

0.01
0.008
0.006
0.004
0.002
0

1000

500

1500

Time (s)
Fig. 5. Model estimates vs. in-field measurements.

Table 6
R2 values and slopes of measurements vs. estimates regression line.
Bus series

19XX
62XX
630X
632X

VT-CPFM

CMEM

MOVES

R2

Slope

R2

Slope

R2

Slope

0.82
0.80
0.76
0.80

0.81
0.88
0.75
0.85

0.81
0.80
0.72
0.80

0.66
0.76
0.73
0.69

0.74
0.74
0.68
0.70

0.82
1.20
1.06
0.60

of a2 is basically equal to 1E07, implying that a value of 1E07


may ensure that the model is not affine to vehicle power and that
the bus optimum fuel economy cruise speed is in a typical range
which is to be determined in Section 5.
Fig. 2 compares the models between bus series. The functional
forms for 19XX- and 632X-series differ significantly from those
for 62XX- and 630X-series. This may be attributed to the differences in vehicle model year, engine and transmission model and
bus empty weight. Although the engine and transmission parameters of 632X series are identical to those of 62XX- and 630X-series,
as described in Table 1, the buses of 632X series weigh much more
than those of 62XX- and 630X-series, which results in higher fuel
consumption levels for buses of 632X series to achieve the identical vehicle power. The 62XX- and 630X-series have the similar
functional trends since their vehicle-specific properties are identical. Counter intuitively, however, the model of 19XX series does
not differ significantly from that of 632X series, although they capture distinct vehicle parameters (as shown in Table 1). This may
attribute that the buses of 19XX series are older (make year:
2009) than those of 632X series (make year: 2013) and may produce higher level of fuel consumption rates under the same driving
condition; while the buses of 632X series, on the other hand, weigh

more than those of 19XX series, which may also cause extra fuel
consumed; the offset in between might probably cause the similar
fuel trends. In short, bus type significantly affects the feature of the
fuel consumption model.
5. Model validation
The model was validated at an instantaneous fuel consumption
level as well as the level of the optimum cruise speed. The model
performance was also compared with CMEM and MOVES in order
to demonstrate the improvements of the proposed models.
5.1. Instantaneous fuel consumption validation
The model estimates were compared against the field measurements at an instantaneous level in order to evaluate the goodness
of the model fit. To capture the auto-correlation in the fuel consumption level, an exponential smoothing filter was applied. The
smoothing process was modeled using a smoothing factor a% multiplied by the new estimate and 1  a% multiplied by the previously smoothed estimate, as demonstrated in Eq. (15). The
smoothing parameter is optimized by minimizing the sum of
squared error between the fuel estimates and the measurements.
Based on experimentation with 14 buses, a smoothing factor of
0.1 was applied.

st axt 1  ast1

15

Here st is the smoothed fuel estimate at time t; st1 is the previously smoothed fuel estimate, xt is the current fuel estimate, and
a is the smoothing factor.
Fig. 3 gives the example results of model validation (other
results are not shown due to space limitations.), demonstrating

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402

that the series model estimates are typically consistent with field
measurements. Although the models either overestimate or underestimate at some data points, in general the predicted fuel consumption rates follow the peaks and valleys of the measured data.
To statistically evaluate the performance of series models, the
coefficients of determination were computed, as illustrated in
Table 4. It should be noted that the R2 values of individual models
are also provided in order to capture the applicability of series
models to each individual bus by comparing the model performance. Basically, the series models have a good performance in
terms of R2 values ranging between 0.70 and 0.85. Individual models also have similar performance by having R2 values within the
same range (0.700.85). Consequently, calibrating the model for
each individual bus does not enhance the model performance significantly. This suggests that the model be calibrated based on the
entire bus series in order to make the modeling practice costeffective and simple.

401

the bus fuel consumption modeling practice. A regression-based


procedure is designed to calibrate the model due to a lack of publicly available fuel data on buses. The results demonstrate that the
bus fuel consumption functional form is concave instead of convex
as was the case with LDVs. The model estimates are demonstrated
to be consistent with in-field measurements. Furthermore, the
model calibrated for an entire bus series is applicable to accurately
capture the fuel consumption behavior of each individual bus
within that series. The optimum fuel economy cruise speeds range
between 40 and 50 km/h which is lower than that for LDVs (60
80 km/h). The model is demonstrated to capture transient fuel consumption behavior better than MOVES and produces a better
agreement to field measurements compared to CMEM.
It is recommended that EPA requires transit buses to report
their fuel economy in the future so that the model can be calibrated using publicly available data without the need for field data
collection.

5.2. Optimum cruise speed

Acknowledgments

Optimum fuel economy cruise speed can be achieved when the


bus fuel consumption rate reaches the minimum level. In validating the series models, the fuel consumed per travel distance was
curved as a function of cruise speed, as illustrated in Fig. 4. Each
series model produces the same bowl shaped curve which is consistent with the prediction of optimum speeds by Rakha et al.
[24] in modeling LDV fuel consumption. Specifically, Table 5 presents the optimum speeds for both series- and individualmodels, ranging between 40 and 50 km/h which is lower than
LDVs (6080 km/h).

This research was sponsored by Blacksburg Transit and the


TranLIVE University Transportation Center (UTC). The authors
would also like to acknowledge all the personnel who provided
technical support or assistance in the data collection and processing activities.

5.3. Model evaluation and implementation


To further evaluate the performance of the proposed model, the
model estimates were compared to CMEM and MOVES. CMEM and
MOVES were selected given that they are two of the most commonly used fuel consumption models in the United States. Fig. 5
illustrates the model estimates superimposed on the field measurements (due to space limitations, only two sample results are
presented in the paper). In general, the estimates of the three models replicate the peaks and valleys in the empirical data, producing
a good agreement with the in-field measurements. Table 6 summarizes the performance of each model. VT-CPFM and CMEM basically produce better estimates by achieving higher R2 values
compared to MOVES, given that MOVES is designed for conformity
use instead of instantaneous analysis. In addition, based on the
slopes of the regression lines between model estimates and field
measurements, the VT-CPFM and CMEM tend to underestimate
the fuel consumption levels (slopes are smaller than 1.0); whereas
VT-CPFM provides better approximation to field measurements by
having the slopes closer to 1.0 compared to CMEM.
The validated models can be implemented either to develop
eco-routing or eco-driving system or to evaluate the effectiveness
of fuel-related techniques and traffic operational strategies in
reducing bus fuel consumption levels. For instance, the models
have been applied to develop an on-demand transit bus ecorouting system in Blacksburg, Virginia. For more details of the system, see [39]. Furthermore, the results of this study are also used
by [40] to quantify the hybrid-induced fuel benefit.
6. Conclusions and recommendations for further research
The research presented in this paper extends the VT-CPFM
framework to model conventional diesel bus fuel consumption
levels in order to circumvent the bangbang type of control in

References
[1] Greene DL, Plotkin S. Reducing greenhouse gas emission from US
transportation. Arlington: PEW Center on Global Climate Change; 2011.
[2] Boriboonsomsin K, Barth MJ, Zhu W, Vu A. Eco-routing navigation system
based on multisource historical and real-time traffic information. IEEE Trans
Intell Transp Syst 2012;13(4):1694704.
[3] Rakha H, Ahn K, Moran K. Integration framework for modeling eco-routing
strategies: logic and preliminary results. Int J Transp Sci Technol 2012;1
(3):25974.
[4] Ahn K, Rakha HA. Network-wide impacts of eco-routing strategies: a largescale case study. Transp Res Part D 2013;25:11930.
[5] Schall DL, Mohnen A. Incentivizing energy-efficient behavior at work: An
empirical investigation using a natural field experiment on eco-driving. Appl
Energy 2015. http://dx.doi.org/10.1016/j.apenergy.2015.10.163.
[6] Saboohi Y, Farzaneh H. Model for developing an eco-driving strategy of a
passenger vehicle based on the least fuel consumption. Appl Energy 2009;86
(10):192532.
[7] Ahn K, Rakha H, Moran K. ECO-cruise control: feasibility and initial testing. In:
Transportation research board 90th annual meeting. No. 11-1031; 2011.
[8] Soylu S. The effects of urban driving conditions on the operating characteristics
of conventional and hybrid electric city buses. Appl Energy 2014;135:47282.
[9] Wayne WS, Clark NN, Nine RD, Elefante D. A comparison of emissions and fuel
economy from hybrid-electric and conventional-drive transit buses. Energy
Fuels 2004;18(1):25770.
[10] Frey HC, Rouphail NM, Zhai H, Farias TL, Gonalves GA. Comparing real-world
fuel consumption for diesel-and hydrogen-fueled transit buses and
implication for emissions. Transp Res Part D 2007;12(4):28191.
[11] Guo J, Ge Y, Hao L, et al. Comparison of real-world fuel economy and emissions
from parallel hybrid and conventional diesel buses fitted with selective
catalytic reduction systems. Appl Energy 2015;159:43341.
[12] Rakopoulos DC, Rakopoulos CD, Giakoumis EG. Impact of properties of
vegetable oil, bio-diesel, ethanol and n-butanol on the combustion and
emissions of turbocharged HDDI diesel engine operating under steady and
transient conditions. Fuel 2015;156:119.
[13] Cook R, Glover EL. Technical description of the toxics module for MOBILE 6. 2
and guidance on its use for emission inventory preparation. Ann Arbor (MI):
Office of Transportation and Air Quality, US EPA; 2002.
[14] EMFAC2007. 2.30 user guide: calculating emission inventories for vehicles in
California. California Air Resource Board; 2007.
[15] Rakha H, Yue H, Dion F. VT-Meso model framework for estimating hotstabilized light-duty vehicle fuel consumption and emission rates. Can J Civ
Eng 2011;38(11):127486. http://dx.doi.org/10.1139/l11-086.
[16] Barth M, An F, Younglove T, et al. Comprehensive modal emission model
(CMEM), version 2.0 users guide. University of California, Riverside; 2000.
[17] MOVES2010. User guide for moves2010a. Tech rep. EPA-420-B-10-036. US
Environmental Protection; 2010.
[18] Salwin AE. Key findings from the intelligent transportation systems program:
what have we learned. McLean (Virginia): Mitretek Systems, Inc. Publication
MP 96W0000195; 1996.

402

J. Wang, H.A. Rakha / Applied Energy 170 (2016) 394402

[19] Barnett L. Netsim users manual. University of Richmond Department of Math


and Computer Science technical report TR-92-01; 1991.
[20] Smit R, Smokers R, Rab E. A new modelling approach for road traffic
emissions: VERSIT+. Transp Res Part D 2007;12(6):41422.
[21] Hausberger S, Rexeis M, Zallinger M, Luz R. PHEM user guide for version 10.
TUG/FVT report; 2010. p. 157.
[22] Cappiello A, Chabini I, Nam EK, Lue A, Abou Zeid MA. A statistical model of
vehicle emissions and fuel consumption. In: Proceedings. The IEEE 5th
international conference on intelligent transportation systems, 2002. IEEE;
2002. p. 8019.
[23] Rakha H, Ahn K, Trani A. Development of VT-Micro model for estimating hot
stabilized light duty vehicle and truck emissions. Transp Res Part D 2004;9
(1):4974.
[24] Rakha HA, Ahn K, Moran K, Saerens B, Bulck EVd. Virginia tech comprehensive
power-based fuel consumption model: model development and testing.
Transp Res Part D 2011;16(7):492503.
[25] Fiori C, Ahn K, Rakha HA. Power-based electric vehicle energy consumption
model: model development and validation. Appl Energy 2016;168:25768.
http://dx.doi.org/10.1016/j.apenergy.2016.01.097.
<http://
www.sciencedirect.com/science/article/pii/S030626191630085X>.
[26] Ahn K, Rakha H. The effects of route choice decisions on vehicle energy
consumption and emissions. Transp Res Part D 2008;13(3):15167.
[27] Barth M, Scora G, Younglove T. Modal emissions model for heavy-duty diesel
vehicles. Transp Res Rec: J Transp Res Board 2004;1880(1):1020.
[28] North RJ. Assessment of real-world pollutant emissions from a light-duty
diesel vehicle. Ph.D. thesis. Imperial College London; 2007.
[29] Nam EK. Proof of concept investigation for the physical emission rate
estimator (PERE) to be used in MOVES. Assessment and Standards Division
Office of Transportation and Air Quality, US EPA; 2003.

[30] Faris WF, Rakha HA, Kafafy RI, Idres M, Elmoselhy S. Vehicle fuel consumption
and emission modelling: an in-depth literature review. Int J Veh Syst Model
Test 2011;6(34):31895.
[31] Guo C, Yang B, Andersen O, Jensen CS, Torp K. EcoMark 2.0: empowering ecorouting with vehicular environmental models and actual vehicle fuel
consumption data. GeoInformatica 2014:133.
[32] Pontryagin LS. Mathematical theory of optimal processes. CRC Press; 1987.
[33] Saerens B, Diehl M, Van den Bulck E. Optimal control using Pontryagins
maximum principle and dynamic programming. In: Automotive model
predictive control. Berlin, Heidelberg: Springer Verlag; 2010. p. 11938.
[34] Park S, Rakha HA, Ahn K, Moran K. Virginia Tech comprehensive power-based
fuel consumption model (VT-CPFM): model validation and calibration
considerations. Int J Transp Sci Technol 2013;2(4):31736.
[35] Wong JY. Theory of ground vehicles. Ottawa (Canada): John Wiley & Sons;
2001.
[36] Feng C. Transit bus load-based modal emission rate model development. Ph.D.
thesis. Georgia Institute of Technology; 2007.
[37] Rakha H, Lucic I, Demarchi SH, Setti JR, Aerde MV. Vehicle dynamics model for
predicting maximum truck acceleration levels. J Transp Eng 2001;127
(5):41825.
[38] Edwardes WA. Modeling diesel bus fuel consumption and dynamically
optimizing bus scheduling efficiency. Ph.D. thesis. Virginia Polytechnic
Institute and State University; 2014.
[39] Witten T, et al. Transit bus routing on-demand: developing an energy-saving
system. Blacksburg Transit, No. 0704-0188, Blacksburg, VA; 2015.
[40] Wang J, Rakha HA. Hybrid-electric bus fuel consumption modeling: model
development and comparison to conventional buses. In: Transportation
research board 95th annual meeting. no. 16-0660; 2016.

S-ar putea să vă placă și