CI9-T-16 Transport and Environment
Modelling instantaneous vehicle emissionsMarc Stettler, Rosalind
ODriscoll, Helen ApSimon, Simon Hu, Jiahui Yang, Yiheng Guo, Justin
Bishop, Adam Boies, Nick Molden
[email protected] |
www.imperial.ac.uk/people/m.stettlerCentre for Transport Studies |
Department of Civil and Environmental Engineering Imperial College
London@TransEnvLab_IC
6th April 2017Institute of Air Quality Management DMUG 2017
OutlineMotivation
Overview of road transport emissions models
Introduction to PEMS data
New emissions models Emissions maps from PEMSNeural networks
Preliminary application
2
MotivationUrban air pollution challenges8% of Europeans1 exposed
to harmful levels of NO2Major contribution from transportUK cities
required to bring air quality into compliance with regulations
Shortcomings of standard emissions models Uncertainty propagates
to forecasts of urban air qualityLimited high temporal &
spatial resolution modelling Limited detail in emissions mechanisms
and chemistry
Real-world diesel emissions are around 5 (1-22) times higher
than RDE limit from Oct3
esa.int
1EEA, Air Quality in Europe 2014 Report, EEA Report No
5/2014European Environment Agency, Copenhagen, Denmark (2014)
18 exceedances of 200gm3 NO23
Air quality (NO2) in LondonEURO 3EURO 4EURO 5EURO 64NO2
concentrations have not improved even though vehicle emissions
standards have become stricter
Road transport emissions modelsAverage speed COPERT
Vehicle specific powerEPA Motor Vehicle Emissions Simulator
(MOVES)International Vehicle Emissions (IVE) modelAlso used for
microscopic prediction
Drive cycle (traffic) parametersVERTSIT+ (TNO)EnViVer plug-in
for PTV VissimHBEFA
Engine/emissions mapsPHEM (TU Graz)AIRE plug-in for
S-Paramics
Physics-basedComprehensive Model Emissions Model (CMEM)CMEM
plug-in for Paramics
Function of vehicle speed and acceleration - f(v,a)Luc Int Panis
et al. (2006)
Black-boxNeural networks
5Street-level and upMicroscopic (instantaneous)
5
Engine load6
FliftFpropulsionFweightFrollFdragFnormalFpropulsion = Fdrag +
Froll + Fgrade + ma
FgradeThe propulsive force is equal to the sum of resistive,
gradient and inertial forces:F = force (N)m = mass (kg)a =
acceleration (m/s2)Fgrade = mg sin
Average speed emissions: COPERT7Emissions are predicted for
different vehicle speed (v) using different functions based on fits
to experimental drive cycle emissions data:
http://emisia.com/products/coperthttp://www.eea.europa.eu/publications/emep-eea-guidebook-2016
Street-level and up
COPERT uncertainties: emissions variability
8http://emisia.com/sites/default/files/COPERT_uncertainty.pdf
Street-level and up
Vehicle specific power: MOVES (EPA)Power = Force velocityVehicle
specific power (VSP) is the engine power divided by the vehicle
mass9
https://www.epa.gov/moves Street-level and up
Statistical regression: VERSIT+ (TNO)10
http://www.sciencedirect.com/science/article/pii/S1361920907000521
Street-level and up
Engine maps: PHEM (TU Graz)
11http://www.ivt.tugraz.at/index.php?option=com_content&view=article&id=69&Itemid=301&lang=en
Microscopic
Physics-based: CMEM (UC Riverside)
12http://www.cert.ucr.edu/cmem/docs/CMEM_User_Guide_v3.01d.pdf
Microscopic
Correlation to speed and acceleration: f(v,a)The emissions rate
(ER) is a function of vehicle speed (v) and acceleration (a)
f(v,a)Calibration parameters (f) that are specific to each
vehicle13Int Panis et al. (2006) Modelling Instantaneous Traffic
Emission and the Influence of Traffic Speed Limits. Science of The
Total Environment 371(1): 27085
http://www.sciencedirect.com/science/article/pii/S004896970600636X
Microscopic
Neural networksBlack-box approach (model structure is
learned)Non-linear NOx formation processesEngine emissions and
effectiveness of a series of emissions control devices is highly
complexExhaust gas re-circulationDiesel oxidation catalystDiesel
particulate filterNOx control (selective catalytic reduction or
lean NOx trap)Ammonia slip
catalyst14Inputs:SpeedAccelerationVSPRPMExhaust
temperatureOutput:NOx (g/s)Fuel consumption
Microscopic
Why do we need to develop more models?Compare approaches and
continuously improve air quality impact estimates
Latest real-world emissions test data indicates:Diesel Euro 6
passenger car emissions are ~5 times higher than limitSignificant
variability between manufacturer/modelDiscrepancy between
laboratory and track/road testingNOx emissions sensitive
toAccelerationAmbient temperatureEmissions control technology
state
Alternative vehicles/powertrainsHybrid electric vehiclesRange
extended electric vehiclesConnected and autonomous vehicles
15
Opportunity to use PEMS dataOn-road emissions for >1000
vehicles, 2-3 new vehicles added each week
Use real-world emissions data to develop instantaneous
models:Extract emissions maps from PEMS real-world emissions data
(1 Hz) (i.e. similar to PHEM)Use a neural network technique and
evaluate different inputs and data processingDemonstrate use of
instantaneous emissions models to evaluate the emissions benefits
of connected and autonomous vehicles
16
PEMS dataDataset includes Euro 5 and 6 petrol and diesel
vehicles
Mixed motorway/non-motorway route~80 km distance~9100 s
duration
Sensors Inc Semtech-DS PEMS with GPS unitNOx measured by
non-dispersive UV17
ODriscoll et al. 2016. Atmospheric Environment 145:
8191.http://www.sciencedirect.com/science/article/pii/S135223101630721X
Vehicle speed and NOx emissions profile18
ODriscoll et al. 2016. Atmospheric Environment 145:
8191.http://www.sciencedirect.com/science/article/pii/S135223101630721X
Real-world emissions complianceData below is for 39 diesel Euro
6 vehiclesNot-to-exceed (NTE) limits for Euro 6c (RDE)2.1 (0.168
g/km) from Sep 20171.5 (0.12 g/km) from 2020A few vehicles with
different emissions control devices already complyLNT (L), SCR
(S)
19
ODriscoll et al. 2016. Atmospheric Environment 145:
8191.http://www.sciencedirect.com/science/article/pii/S135223101630721X
www.equaindex.com
Comparison to COPERTCOPERT underestimates compared to the
average over all vehicles measured in real-world over the entire
cycleMeasured NOx and NO2 are 1.6 and 2.5 times higher than
COPERT20
ODriscoll et al. 2016. Atmospheric Environment 145:
8191.http://www.sciencedirect.com/science/article/pii/S135223101630721X
Approach: Bottom-up emissions modellingEmissions map models
relies on extracting lookup tables of emissions (i.e. emissions as
a function of engine speed and torque):
Use PEMS measurements and on-board diagnostics (OBD) data (e.g.
engine speed) to extract effective gear ratios in order to
calculate engine torque.
Extract the emissions measurements for a given engine speed and
torque range.
Simulate a given drive cycle (vehicle speed versus time)Bishop,
et al. 2016. Applied Energy 183:
20217.http://www.sciencedirect.com/science/article/pii/S0306261916312843
21
JB: Add flow schematic from last year and citation, send last
years presentation21
1. Extracting gear ratios (e.g. for truck)22
Bishop, et al. 2016. Applied Energy 183:
20217.http://www.sciencedirect.com/science/article/pii/S0306261916312843
2. Extracting emissions maps23
Bishop, et al. 2016. Applied Energy 183:
20217.http://www.sciencedirect.com/science/article/pii/S0306261916312843
2. Extracting emissions maps (passenger cars)
log(NOx) in g/slog(NOx) in g/slog(NOx) in g/slog(NOx) in
g/s24Bishop et al., (2017). Int. J. Transp. Dev. Integr., Vol. 1,
No. 2 (2017)
JB: Units on scale and have logarithmic if possibleLog scale
added, values are negative because they are less than 124
3. Transient emissions prediction (1Hz)
25Bishop et al., (2017). Int. J. Transp. Dev. Integr., Vol. 1,
No. 2 (2017)
Remove fuel25
Neural network modelEvaluate:Different ways to define the
training data setDifferent input dataSpeedAccelerationVSPRPMExhaust
temperature
Number of hidden layers and data averaging not
discussed26Output:NOx (g/s)Fuel consumption
Preliminary please do not cite or quote
Inputs from PEMS data
Training data set sampling27
Sequential (default)
Random selection of data blocks (5 seconds)
Training(65%)Validation (15%)Testing (20%)Preliminary please do
not cite or quote
27
Accuracy depends on input data to NNScenarioVehicle
speedVSPAccelerationEngine speedExhaust
temperatureR210.19-0.6820.43-0.6530.47-0.7240.48-0.7350.66-0.8560.65-0.84
28
Preliminary please do not cite or quote
Comparison of NN and f(v,a)ANN: neural network calibrated to
each vehicleSelf-calibrated: f(v,a) calibrated to individual
vehicleCalibrated for all: f(v,a) calibrated to combination of five
vehiclesPanis 2006: f(v,a) same as from Int Panis et al.
(2006)29
Preliminary please do not cite or quote
What if we only have vehicle speed and acceleration data?
Comparison of NN and f(v,a)
30Preliminary please do not cite or quote
Accuracy is currently inconsistent
31
Preliminary please do not cite or quote
Summary and future workInstantaneous emissions modelling for NOx
Better able to capture high emissions events but missing the
peaksNeeds further validation and improvement compared to
real-world PEMS data
Challenges:Do we need emissions data for each vehicle on the
road, and is this feasible?Can we obtain accurate instantaneous
vehicle trajectories (and other data) for each vehicle?How to
account for ambient temperature effects and cold startPrimary NO2,
PM, NH3, N2O
In order to estimate air quality at the city scale, a hybrid
approach using data from GPS, ANPR, traffic cameras and remote
sensing is likely to be required32
What is a CAV?Connected Autonomous Vehicle can communicate with
other vehicles (V2V)
Vehicle can communicate with road infrastructure (V2I)Vehicles
have different longitudinal behaviour
Vehicles have different lateral behaviour
Vehicles have better throttle control(Preliminary) Emissions
benefits of CAVs
CAV zero emissions
33
Optimised Vehicle Autonomy for Ride and Emissions (OVARE)
CAV modelling V2V vehicle to vehicleV2I vehicle to
infrastructure
34
Traffic simulation detailsVISSIM South Kensington traffic
modelModelled period: 7:30 to 9:30 AM peak modelTotal number of
links: 333Number of junctions: 20Each simulation run: 20 minutes (@
CPU:3.1 GHz, RAM: 8 GB)
Calibration of the modelTraffic flow is calibrated against
survey data collected at key locations on the networkRouting choice
is calibrated based on O-D surveySignal timing is used for existing
fixed timing
Instantaneous emissions modelUse f(v,a) calibrated to one
vehicle (for preliminary analysis)
35Preliminary please do not cite or quote
CAV simulation around South Kensington36
Preliminary please do not cite or quote
0% CAV100% CAV37(Preliminary) emissions benefits of
CAVsPreliminary please do not cite or quote
Thank you!Marc Stettler, Rosalind ODriscoll, Helen ApSimon,
Simon Hu, Jiahui Yang, Yiheng Guo, Justin Bishop, Adam Boies, Nick
Molden
[email protected] |
www.imperial.ac.uk/people/m.stettler@TransEnvLab_ICCentre for
Transport Studies | Department of Civil and Environmental
Engineering Imperial College London
6th April 2017Institute of Air Quality Management DMUG 2017