Professional Documents
Culture Documents
The Application of The CO2MPAS Model For Vehicle CO2 Emissions Estimation Over Real Traffic Conditions
The Application of The CO2MPAS Model For Vehicle CO2 Emissions Estimation Over Real Traffic Conditions
The Application of The CO2MPAS Model For Vehicle CO2 Emissions Estimation Over Real Traffic Conditions
Transport Policy
journal homepage: http://www.elsevier.com/locate/tranpol
A R T I C L E I N F O A B S T R A C T
Keywords: A vehicle simulation model, CO2MPAS supports the introduction of the new WLTP-based certification system for
Light duty vehicles CO2 emissions in Europe. This paper investigates the possibility to use the underlying simulation methodology to
CO2 emissions accurately calculate CO2 emissions over real-world trips, thus to extend the use of the methodology beyond
CO2MPAS
vehicle certification. As a reference, the analysis used measurement data obtained from four vehicles over two
EMEP/EEA
Real-world
different routes under real-world driving conditions. The CO2 emissions were measured using portable emissions
Fuel consumption gap measurement systems. The formal CO2MPAS methodology and two modified versions of it that require a reduced
number of input data were assessed about their capacity to predict the measured CO2 emissions. The analysis
focused on the accuracy and uncertainty of the three different methodology configurations. As an additional
benchmark, the analysis considered the CO2 emissions estimates obtained from the EMEP/EEA Guidebook
methodology used for emissions inventorying in Europe. Results show that the basic CO2MPAS configuration
demonstrates good performance in predicting CO2 emissions over on-road tests, reaching a prediction accuracy
over an entire test trip of 0,3% and a standard deviation of 3,1%. The modified versions showed slightly higher
biases up to 3% and uncertainties (5–7%), but remaining within reasonable limits considering the reduced
number of inputs used in each case. Given its ability to predict CO2 emissions accurately on a local base,
CO2MPAS could be used for the prediction of instantaneous CO2 emissions in traffic micro-simulation exercises.
☆
The views expressed are purely those of the authors and may not in any circumstance be regarded as an official position of the European Commission.
* Corresponding author.
E-mail address: georgios.fontaras@ec.europa.eu (G. Fontaras).
https://doi.org/10.1016/j.tranpol.2020.01.005
Received 19 October 2019; Accepted 8 January 2020
Available online 13 January 2020
0967-070X/© 2020 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
C. Mogno et al. Transport Policy 124 (2022) 152–159
153
C. Mogno et al. Transport Policy 124 (2022) 152–159
approval conditions. In addition to the chassis dyno test, for each vehicle The basic configuration operates in the same way as over the certi
two real-world measurements were performed. The two real-world fication process. Data originating from a WLTP-test are used to calibrate
driving routes and cycles followed in all the tests are described in the model. In this study CO2MPAS physical model was calibrated and
(Valverde Morales and Bonnel, 2018). CO2 emissions over real world optimized by using the WLTP test-data for each of the vehicles selected.
conditions were measured according to the general guidelines of the EU Subsequently, instead of predicting NEDC CO2 emissions, CO2MPAS was
regulation, with standard PEMS equipment, the characteristics of which used to predict the CO2 emissions over the real world tests. CO2MPAS-
are described in (AVL M.O.V.E GAS PEMS IS) and in (AVL M.O.V.E EFM generic configuration uses to some extent the same input data as the
Exhaust Flow Meter). PEMS instrumentation has shown significant po CO2MPAS basic but doesn’t undergo the calibration process for each
tential in a variety of studies related to real-world emissions: in the specific vehicle. Instead, generic empirical models are used to model the
direct analysis of the gaseous emissions from light-duty diesel vehicle operation of vehicle components such as the engine or the gearbox
(Luja�n et al., 2018; Mahesh et al., 2018), in the assessment of the impact (Tsiakmakis et al., 2017). The usefulness of CO2MPAS generic becomes
of different driving styles and route characteristics on on-road exhaust evident in cases of studies were only limited information is available
emissions (Gallus et al., 2017), and in the development and validation of regarding the vehicles being simulated. Finally, the raster approach is a
passenger car emission factors (Kousoulidou et al., 2013). All mea “downsized” version of the CO2MPAS-generic methodology, where the
surements were acquired at a frequency of 1 Hz. The main characteris fuel consumption is calculated using a two-dimensional function/table
tics of the four tested vehicles can be found in Table 1. (raster) of velocity and power. The raster can be either calibrated using
CO2MPAS, or receive input by the user based on experimental data or
2.2. CO2 emissions models other sources. This methodology has been implemented in the JRC’s
U-SAVE tool, a software customized for optimal fuel consumption
Two emissions calculation models were chosen for reproducing the routing and velocity optimization (Arcidiacono et al., 2017). The raster
measured CO2 and fuel consumption in this exercise. The CO2MPAS configuration of CO2MPAS offers improved characteristics with respect
model developed by the JRC in the framework of the EU regulations to code execution speed and adaptability, features that are necessary for
2017/1152/EU and 2017/1153/EU, and the EMEP/EEA methodology routing and path optimization applications. All three CO2MPAS con
which is used by most EU members states for emissions inventorying and figurations take as input from the real world tests, velocity time-series
projections purposes. As mentioned in the introduction, CO2MPAS is a and the road grade. Original data resolution of 1 Hz has been kept for
vehicle CO2 emissions and energy consumption simulator, created to the calculation of CO2 emissions.
facilitate the introduction of the WLTP test protocol in the European The most known software implementation of the EMEP/EEA
legislation. The core of CO2MPAS is a physical model, backward- Guidebook methodology for the calculation of air pollutant emissions
looking, longitudinal dynamics simulator that calculates the energy from road transport is COPERT that is used by many EU states for
flow and energy losses at various vehicle-components. CO2MPAS uses as emissions inventorying purposes and by individual users for scientific
input data from the WLTP certification test combined with NEDC rele applications and academic research (EMISIA, 2018; European Environ
vant parameters of the vehicle (e.g. mass, road loads and test boundary ment Agency, 2016; Saharidis and Konstantzos, 2018, Panis et al.,
conditions) in order to simulate vehicle CO2 emissions and fuel con 2011). According to the EMEP/EEA approach, energy consumption is
sumption. Data acquired over the WLTP are used for calibrating and calculated based on the average speed of the vehicles and of a series of
optimizing the physical sub-models comprising CO2MPAS, while the tabulated coefficients that take into account the vehicle class, fuel type,
final model configuration for the specific vehicle is used to predict the and engine technology; the main advantage of this is that its method
NEDC CO2 emissions, or in the case of the present study over the driving ology can achieve detailed emissions calculations with the use of few
cycle/driving route in investigation (Fontaras et al., 2018). CO2MPAS is input data (Kouridis et al., 2010). Subsequently the energy consumption
open source tool, and it is available online. Three different variations of is linked to vehicle type and fuel characteristics in order to calculate the
the CO2MPAS methodology were investigate in this study: resulting tailpipe CO2 emissions.
In this exercise, the latest version of the methodology was used, as
� the basic configuration (CO2MPAS-basic) (Fontaras et al., 2018) described in (European Environment Agency, 2016). In further detail,
� a generic version (CO2MPAS-generic) similar to the approach pre following the EMEP/EEA approach, as implemented in COPERT version
sented by (Tsiakmakis et al., 2017) 5, first the energy consumption was calculated through the velocity
� a version where fuel consumption is simulated through a bi- time-series and the travelled distance, then the corresponding CO2
dimensional matrix (raster approach) (Arcidiacono et al., 2017). emissions are derived after the fuel consumption calculation. Only hot
emissions were considered in the study, because the cold start emissions
Table 1
Main characteristics of selected vehicles.
Vehicle 1 Vehicle 2 Vehicle 3 Vehicle 4
Body type 4/5seater hatchback 4/5seater suv/sports 4/5seater hatchback 4/5seater hatchback
Engine 3 cyl turbocharged gasoline 4 cyl turbocharged diesel with common rail 4 cyl gasoline 4 cyl turbocharged diesel with common rail
Capacity 999 cc 1956 cc 1242 cc 1968 cc
Max Power 70 kW 103 kW 51 kW 110 kW
@5000–5500 RPM @ 4000 RPM @ 5500 RPM @ 3500 RPM
Mass (MRO)a 1453 kg 1789 kg 1253 kg 1719 kg
Max Torque 160 Nm 350 Nm 102 Nm 320 Nm
@ 1500–3500 RPM @ 1750 RPM @ 3000 RPM @ 1750–3000 RPM
Max Speed 186 km/h 190 km/h 164 km/h 214 km/h
Drive Front All front Front
Gearbox 5 speed manual 9 speed automatic 5 speed manual 6 speed manual
Wheels 215/40R17 225/45R18 175/65R14 225/40R18
Euro-class EURO-6 EURO-6 EURO-6 EURO-6
Type approval CO2 98 g/km 144 g/km 119 g/km 117 g/km
Test load 140 Kg 110 kg 140 kg 140 Kg
a
For this exercise, Mass in Running Order MRO ¼ Empty weight of the vehicle þ 75 � 2 drivers weight þ 85% fuel tank capacity þ PEMS test load.
154
C. Mogno et al. Transport Policy 124 (2022) 152–159
were calculated to be of marginal impact, given the length of the test trip CO2MPAS-raster. However, these two CO2MPAS versions exhibit a
without significant stops (less than 1% of the total CO2 over each trip). similar behavior considering the trend in the predictions for each vehicle
Furthermore, an additive correction in the fuel consumption (and and test trip. Finally, EMEP/EEA appears to have a bias of the order of
consequently, in the CO2 emissions) proposed by (Ntziachristos et al., 10%, and a standard deviation of the order of 7%, a value comparable
2014) was implemented. This correction takes into account the type to that of CO2MPAS raster. EMEP/EEA shows a tendency to under-
approval fuel consumption, engine capacity, and vehicle effective mass. predict the total CO2 emissions. For the three vehicles, the EMEP/EEA
Moreover, the overall procedure described above has been applied on model always under-estimates the total CO2 emissions, within a range of
averaged velocity time series per each kilometer travelled, in order to values between 8% and - 18%. Only for one vehicle out of four (Vehicle
match the suitable spatial and temporal resolution for using this 3) there are slight over-predictions, with values below the 1%. However,
methodology. this observation could be case-specific. , Global emission models are in
general calibrated to provide minimum bias over an entire fleet and
3. Results aggregate annual operating conditions. This is the most interesting case
as results for Vehicle 3 seem to contradict the overall considerations
In this section, the results of CO2MPAS prediction for emissions from expressed above. Indeed, the best total CO2 predictions for Vehicle 3 are
the real world tests (RWT) performed are analyzed and compared to the given by the EMEP/EEA model followed by CO2MPAS generic, CO2M
measured CO2 value and the one simulated using the EMEP/EEA PAS basic and lastly CO2MPAS raster. This occurs for both tests
methodology. First, the total CO2 predictions over each test trip is dis (demonstrated by the prediction error for the different models for both
cussed. Subsequently, a more detailed analysis of the local CO2 pre Test 1 and Test 2 in Table 2). However, to fully explain these results, it is
dictions over the test trips is provided. necessary to analyze in detail the local CO2 predictions for the four
models (see the next subsection). Finally, it is worth mentioning that, for
all three models tested, no significant correlation between the vehicle
3.1. Global CO2 emissions predictions fuel-type or the test trip number (Test 1 or Test 2) and the total CO2
prediction emerged from the results.
Table 2 summarizes the overall CO2 predictions results for all the
models, by test trip and by vehicle. Overall, CO2MPAS basic emerges to
be the most precise and accurate model for the total CO2 prediction. For 3.2. Local CO2 emissions predictions
the set of vehicles and test trips simulated, CO2MPAS basic shows a
prediction accuracy within a maximum error of 4.4% from the real Table 3 summarizes the average prediction error per km for the three
total CO2 emissions for the single test trip, and it displays a significant configurations of CO2MPAS and for the EMEP/EEA calculations. The
precision with only a 3.1% standard deviation for the entire set of pre results refer to the average of the mean error per km and of the
dictions. Moreover, considering the combined CO2MPAS basic results in respective standard deviation over the eight trips performed. The rela
Table 2, this model doesn’t exhibit any evident bias for under-prediction tive error is considered with respect to the average CO2 emissions per
or for over-prediction of the total real CO2. This is confirmed by the km, calculated from the RWT for the specific vehicle and trip under
average prediction error of only – 0.3%. consideration. From the table it is evident that also locally CO2MPAS
The CO2MPAS generic and the CO2MPAS raster versions give on basic has the best accuracy and precision among the three CO2MPAS
average less accurate and precise results, with an average prediction version, with an average error per km of 0.3% with an average
error of 3.3% and 1.1%, and with a standard deviation of 5.7% and dispersion of 15.2%. It is followed by CO2MPAS generic and CO2MPAS
7.0% respectively. From these results, one could argue that overall the raster that comes out to have a comparable local prediction, with a mean
global accuracy (bias) of all three CO2MPAS configurations is compa error of 3.1% and 0.9% and mean dispersion of 21.5% and 21.2%
rable; however precision clearly degrades, as the inputs are reduced respectively. These figures confirm the tendency of the EMEP method
with CO2MPAS generic presenting a lower uncertainty compared to the ology to underestimate the CO2 emissions of the specific vehicle-trip
combinations also locally. A visual representation of these results is
Table 2 given in Fig. 1 and in Fig. 2. Histograms in Fig. 1 show the distribution of
Prediction error for the total CO2 for each vehicle and test considered. the relative error of each model per km for a selected trip. The vehicle
Test Vehicle Prediction error for the total CO2 (%) and trip combinations selected for each of the models in Fig. 1 are those
that exhibited the average performance closest to the overall average
CO2MPAS CO2MPAS CO2MPAS COPERT
basic generic raster model performance (error and average standard deviation) shown in
Table 3. Comparing the four histograms, it emerges the relative preci
Test Vehicle 0.3 2.8 1.2 14.2
1 1
sion of the models, also from the shape of the distributions variability
Vehicle 3.9 12.2 11.0 18.0 and the range of the relative error.
2 As a complementary example, Fig. 2 shows the comparison of the
Vehicle 3.5 1.3 6.7 0.3 relative error distribution for the detailed CO2 prediction of the four
3
models but over the same trip. From this figure, it is possible to see how
Vehicle 4.4 4.5 3.7 13.9
4 the relative performance of the models highlighted in Fig. 1 applies also
Test Vehicle 0.6 5.3 3.0 13.8 to a single test trip. Fig. 3 shows in more detail the time-series of the
2 1 simulations performed. In particular, it displays the CO2 emissions per
Vehicle 0.8 9.1 8.5 13.2
2
Vehicle 4.1 3.9 8.7 0.7 Table 3
3 Average prediction error per km for the detailed CO2 for each model.
Vehicle 1.3 2.6 3.4 8.1
Prediction error per km for the detailed CO2 (%)
4
COMPAS
2 COMPAS
2 COMPAS
2 EMEP/
Average ¡0.3 ¡3.3 ¡1.1 ¡10.0
basic generic raster EEA
prediction
error Average 0.3 3.1 0.9 12.3
Standard 3.1 5.7 7.0 7.0 standard 15.2 21.5 21.2 36.3
deviation deviation
155
C. Mogno et al. Transport Policy 124 (2022) 152–159
Fig. 1. Relative error distributions for the detailed CO2 prediction of CO2MPAS and EMEP/EEA for the corresponding average performance of: (a) CO2MPAS basic
for Vehicle 2 Test 2; (b) CO2MPAS generic for Vehicle 4 Test 2; (c) CO2MPAS raster for Vehicle 1 Test 2; and (d) EMEP/EEA for Vehicle 4 Test 2.
closest to the values shown in Table 3 (the same trips depicted also in
Fig. 1). The best performance has been chosen as the performance with
the lowest relative error per km and the lowest standard deviation.
Taking into account the average performance, it is evident CO2MPAS
basic’s capacity to match accurately the real CO2 emissions profile, by
achieving good prediction also over steep CO2 emission variations
(Fig. 3 (a)). This could be attributed both to the level of detail of the
general CO2MPAS0 physical model, and CO2MPAS basic’s calibration
capability. In each case CO2MPAS basic was self-calibrated on the
respective vehicle using WLTP test-data. On the other hand, the in
congruity in the CO2 emissions prediction in some parts of the trip,
especially in some central emissions peaks visible in Fig. 3 (a), can be
attributable to different factors:
� Uncertainties in the input data retrieved from the RWT that lead to
making assumptions on variables such as ambient temperature and
road loads. In the WLTP test cycle, the values of these variables are
fixed, but they can differ from the values during the RWT.
� Driver behavior and lack of information about driving conditions e.g.
the real gear shifting during the RWT is different from the one used
Fig. 2. Comparison of the error distribution for the detailed CO2 prediction of
for the WLTP.
the four models over the same trip (Vehicle 2 Test 1).
� Vehicle features such as the Start-stop system that may not function
as expected during a RWT e.g. there can be stops for which the en
each km travelled over the specific test trip. An average performance gine does not turn off (speed equal to zero but engine power non
and the best performance in the CO2 predictions series are presented for zero)
each CO2MPAS version and for the EMEP/EEA methodology. For each � Use of auxiliary systems that consume fuel such as air-conditioning,
model, the average performance has been chosen among the 8 different radio and other car equipment that impact the results, but for which
trips (2 tests x 4 vehicles) as the performance of a single trip that is information was not recorded during the tests.
156
C. Mogno et al. Transport Policy 124 (2022) 152–159
Fig. 3. Average and best performance for the detailed CO2 prediction. CO2MPAS basic: (a) average for Vehicle 2 Test 2, (b) best for Vehicle 1 Test 1, (c) EMEP/EEA:
average for Vehicle 4 Test 2 and (d) best for Vehicle 3 Test 1, (e) CO2MPAS generic: average for Vehicle 4 Test 2, (f) best for Vehicle 3 Test 1, (g) CO2MPAS raster:
average for Vehicle 1 Test 2 and (h) best for Vehicle 1 Test 1(h).
157
C. Mogno et al. Transport Policy 124 (2022) 152–159
If all this extra information was available, CO2MPAS-basic results model (global prediction: mean error 10%, dispersion 7%; local pre
could further improve avoiding uncertainties and the use of qualified diction: mean error per km of 12.3% and dispersion of 36.3%).
assumptions. However, as stated before, even with all these assump CO2MPAS demonstrates good performance in predicting CO2 emissions
tions, CO2MPAS basic is able to predict with high accuracy almost the over on-road tests and could be used beyond its certification purpose.
entire trip. Fig. 3 (b) shows the best performance of CO2MPAS basic in Results also demonstrate the operation principles and structure of
predicting the detailed CO2 for a test trip. It is worth noticing that even the CO2MPAS tool can be used for making accurate projections of the
for an average local prediction performance, a very good result can be emissions of individual vehicles under real world driving. In addition the
achieved on the CO2 prediction over the entire trip. Indeed, for the methodology assessed in this paper can be used for quantifying the
average local performance for CO2MPAS basic (Fig. 3 (a)), the corre contributions of different vehicle-related or exogenous factors in the
sponding total CO2 prediction has an error of 0.8%. This is caused by energy and fuel consumption of vehicles, even when limited information
the cancelling-out of the over and under-predictions for certain peaks of is available. Applying such an approach would offer significant benefits
the trip, as shown in Fig. 3 (a). As a consequence, even if locally the in the control and characterization of the fuel consumption and CO2
prediction is not so accurate, it can be more accurate globally. This emissions gap. The recently adopted EU regulation 2019/631/EU stip
mechanism of compensation is much more evident in the EMEP/EEA ulates that “The Commission should assess how fuel and energy con
detailed CO2 predictions. Fig. 3 (c) and (d) show how EMEP/EEA sumption data may help to ensure that the vehicle CO2 emissions
detailed prediction is less accurate considering the CO2 emissions profile determined in accordance with the WLTP remain representative of real-
when applied for the specific vehicle and trip conditions. world emissions over time for all manufacturers and, more precisely,
Over the entire trip, the compensation between over-predictions and how such data can be used to monitor the gap between laboratory and
under-predictions can result in a very accurate total CO2 prediction, as it real-world CO2 emissions and, where necessary, to prevent this gap from
occurs for the case of Vehicle 3. However, this circumstance is not sys increasing”. Although the core of such an assessment will be based on
tematic, and as a consequence overall the EMEP/EEA approach exhibits real-world operation data collected throughout the vehicle fleet via on-
wider biases. This behavior is due to the fact that the EMEP/EEA model bard fuel consumption measurement systems (OBFCM), it will still be
operates with very few input information, that is usually available to necessary to employ analysis tools and models, and produce the neces
emissions modeleres, and aims to capture with relative accuracy the sary proxy test-samples, for characterizing the observed gap and iden
emissions of the entire fleet over high-mileage, high time-span boundary tifying the most important factors contributing to it. The development
conditions and to a lesser extent the emissions of individual vehicles specifications of CO2MPAS were challenging, as it had to be highly ac
over specific (short) trips. The EMEP/EEA methodology is indeed used curate, exhibit fast operation, and function with a limited number of
mainly in the compilation of regional and national emissions input data. It is therefore of benefit to build on the experience and
inventories. characteristics of the tool and expand it as necessary, in order to cover
Finally, the performance of CO2MPAS generic and CO2MPAS raster future regulatory needs. The CO2MPAS-basic configuration, as pre
ranks in between those CO2MPAS and EMEP/EEA. This is an expected sented in this first proof-of-concept study, can use data available at the
result as both models operate using less input information compared to official type approval test, such as WLTP CO2 emission values and
CO2MPAS but more compared to EMEP/EEA. Fig. 3 (e) and (f) show an vehicle configuration data, to predict the likely CO2 emissions and en
average performance and the best performance for CO2MPAS generic, ergy consumption of specific vehicle pools under different real world
while Fig. 3 (g) and (h) show the same for CO2MPAS raster. Comparing operating conditions. The backbone to support the regulatory frame
qualitatively these figures with Fig. 3 (a) and (b) (CO2MPAS basic) and work for such use exists to a large extent under the current CO2 certi
with Fig. 3 (a) and (b) it is evident that the accuracy and the precision of fication scheme. Such an approach when combined with information on
CO2MPAS generic and CO2MPAS raster are much higher accuracy and the evolution and distribution of different factors affecting the energy
precision, but a bit lower than CO2MPAS basic one. For these two consumption of vehicles (eg distribution of ambient temperatures, oc
CO2MPAS versions, generic and raster, the detailed CO2 prediction cupancy, traffic conditions) could allow the reconstruction of the likely
match qualitatively the real CO2 emissions profile for the vast majority distributions of the real world CO2 emissions gap on a regular basis. This
of each test trip. Moreover, the qualitative behavior in the local pre type of information will be a valuable supplement and will help un
diction follows the same trend, like in the case of the total prediction, as derstand better, and eventually address, the origins and possible future
mentioned in the previous subsection. increases of CO2 emissions in real world operation.
158
C. Mogno et al. Transport Policy 124 (2022) 152–159
159