Portable Emissions Measurement Systems (PEMS) provide direct real-time data on the emissions of an in-use vehicle under a wide range of real-world operating conditions. In recent years PEMS technology has undergone a very rapid evolution and data gathering activities are now widespread, especially in the US and Asia. But, by comparison, methods used to analyze PEMS data have received relatively little refinement over the same period. The reason for this is in part legislative, but it also reflects an issue that is much more fundamental and commonplace – if datasets are large and noisy, someone has to put a lot of work in if you want to get good information out.
Current PEMS data analysis is perhaps most easily considered in terms of its two extremes, total journey and raw data analysis, as most current practices fall into one of these two categories. Here, we considered these and an alternative ‘middle ground’ approach, micro trip analysis. We also look at different micro-trips sampling strategies and some automation procedures for the routine use of such methods on a much wider range of research questions.
www.its.leeds.ac.uk/people/k.ropkins
Vehicle emissions measurement: micro-trip analysis of non-stationary time-series
1. MICRO-TRIP ANALYSIS OF
NON-STATIONARY TIME-SERIES
Karl Ropkins
Institute for Transport Studies, University of Leeds, Leeds, LS2 9JT, UK
Contact k.ropkins@its.leeds.ac.uk
2014 ITS Seminar Series
ITS, University of Leeds, June 18th 2014
5. Non-stationary Time-series
• The data sets discussed are from portable emission
measurement systems (PEMS)
• These are one example of a non-stationary time-series
• Others include:
• Portable activity measurement systems (PAMS)
• (Increasing number of large vehicle fleets)
• Aircraft Infrastructure Management System (AIMS)
• Animal tracking
• Personal GPS and mobile phone movement
6. PEMS ISA Study
One Study for one Vehicle Management system:
• Two vehicles, two fuels types
• One Intelligent Speed Adaptation (ISA) system,
three operating modes (OFF, ADV, VOL)
• Three routes - but not all vehicles on all routes
• One PEMS - but additional logging
7. In terms of data size:
• An individual journal generates 1,000 to 50,000 records
• A study generates 10,000s to 1,000,000s records
(PEMS ISA example: 1, 080,000 records)
• PEMS data archives like those of the EPA, CARB, etc,
include data from 100s of studies and real-world
certification exercises
8. Total Journey Analysis
Comparison of measurements (summed or standardized) on a ‘per
journey basis’
Approach is analogous to conventional vehicle/engine certification
testing
… BUT in the real-world it is crude approach
For all routes in the PEMS ISA study, e.g.:
• We do not see anything significant in total journey data
• BUT that is not really that surprising
• There is HIGH run-to-run variation
• The impact of ISA is expected to be SMALL
9. Raw Data Analysis and Modeling
Analyzing the data at the resolution it was logged at
Approach has the potential to be more informative but analysis is
more labour-intensive
…and more often you are trading uncertainty
for the perception of certainty
10. Raw Data Analysis and Modeling
Data modeling [Analyte]i = te(speedi-n, acceli-n) + … +
te(speedi-m, acceli-m)
11. Raw Data Analysis and Modeling
Data modeling [Analyte]i = te(speedi-n, acceli-n) + … +
te(speedi-m, acceli-m)
Results
ISA Mode Comparison
OFF vs. ADV OFF vs. VOL ADV vs. VOL
OFF ADV VOL difference % p difference % p difference % p
Diesel Mondeo CO2 (g.km
-1
) 258.19 266.57 260.65 8.38 3.04 0.52 2.46 0.93 0.63 5.92 -1.89 0.63
All Routes CO (g.km
-1
) 0.12 0.12 0.12 0.01 4.00 0.61 0.001 1.41 0.49 0.004 -2.34 0.82
HC (g.km-1
) 0.27 0.28 0.27 0.01 3.15 0.70 0.001 0.31 0.59 0.01 -2.61 0.82
NOx (g.km-1
) 0.74 0.77 0.75 0.03 3.40 0.46 0.01 1.05 0.60 0.02 -2.11 0.64
Fuel economy
(km.litres
-1
)
7.75 7.68 7.73 -0.07 -0.14 0.52 -0.02 -0.63 0.54 -0.06 0.84 0.56
Petrol Mondeo CO2 (g.km
-1
) 370.07 388.75 373.98 18.7 4.36 0.66 3.91 0.77 0.57 14.8 -3.15 0.69
All Routes CO (g.km
-1
) 1.38 1.51 1.41 0.13 7.65 0.55 0.03 1.47 0.39 0.10 -5.17 0.71
HC (g.km
-1
) 0.68 0.72 0.69 0.04 4.42 0.65 0.01 0.72 0.43 0.03 -3.25 0.71
NOx (g.km
-1
) 0.25 0.25 0.25 0.01 1.98 0.57 -0.0003 -0.37 0.60 0.01 -2.16 0.64
Fuel economy
(km.litres-1
)
6.41 6.27 6.38 -0.14 -2.46 0.66 -0.03 -0.74 0.57 -0.11 1.86 0.69
12. Raw Data Analysis and Modeling
Data modeling [Analyte]i = te(speedi-n, acceli-n) + … +
te(speedi-m, acceli-m)
Results
ISA Mode Comparison
OFF vs. ADV OFF vs. VOL ADV vs. VOL
OFF ADV VOL difference % p difference % p difference % p
Diesel Mondeo CO2 (g.km
-1
) 258.19 266.57 260.65 8.38 3.04 0.52 2.46 0.93 0.63 5.92 -1.89 0.63
All Routes CO (g.km
-1
) 0.12 0.12 0.12 0.01 4.00 0.61 0.001 1.41 0.49 0.004 -2.34 0.82
HC (g.km-1
) 0.27 0.28 0.27 0.01 3.15 0.70 0.001 0.31 0.59 0.01 -2.61 0.82
NOx (g.km-1
) 0.74 0.77 0.75 0.03 3.40 0.46 0.01 1.05 0.60 0.02 -2.11 0.64
Fuel economy
(km.litres
-1
)
7.75 7.68 7.73 -0.07 -0.14 0.52 -0.02 -0.63 0.54 -0.06 0.84 0.56
Petrol Mondeo CO2 (g.km
-1
) 370.07 388.75 373.98 18.7 4.36 0.66 3.91 0.77 0.57 14.8 -3.15 0.69
All Routes CO (g.km
-1
) 1.38 1.51 1.41 0.13 7.65 0.55 0.03 1.47 0.39 0.10 -5.17 0.71
HC (g.km
-1
) 0.68 0.72 0.69 0.04 4.42 0.65 0.01 0.72 0.43 0.03 -3.25 0.71
NOx (g.km
-1
) 0.25 0.25 0.25 0.01 1.98 0.57 -0.0003 -0.37 0.60 0.01 -2.16 0.64
Fuel economy
(km.litres-1
)
6.41 6.27 6.38 -0.14 -2.46 0.66 -0.03 -0.74 0.57 -0.11 1.86 0.69
Small penalty for using ISA: Emissions +0.5 to +4%
Fuel economy -0.7 to -2.5%
13. Raw Data Analysis and Modeling
Data modeling [Analyte]i = te(speedi-n, acceli-n) + … +
te(speedi-m, acceli-m)
Results
ISA Mode Comparison
OFF vs. ADV OFF vs. VOL ADV vs. VOL
OFF ADV VOL difference % p difference % p difference % p
Diesel Mondeo CO2 (g.km
-1
) 258.19 266.57 260.65 8.38 3.04 0.52 2.46 0.93 0.63 5.92 -1.89 0.63
All Routes CO (g.km
-1
) 0.12 0.12 0.12 0.01 4.00 0.61 0.001 1.41 0.49 0.004 -2.34 0.82
HC (g.km-1
) 0.27 0.28 0.27 0.01 3.15 0.70 0.001 0.31 0.59 0.01 -2.61 0.82
NOx (g.km-1
) 0.74 0.77 0.75 0.03 3.40 0.46 0.01 1.05 0.60 0.02 -2.11 0.64
Fuel economy
(km.litres
-1
)
7.75 7.68 7.73 -0.07 -0.14 0.52 -0.02 -0.63 0.54 -0.06 0.84 0.56
Petrol Mondeo CO2 (g.km
-1
) 370.07 388.75 373.98 18.7 4.36 0.66 3.91 0.77 0.57 14.8 -3.15 0.69
All Routes CO (g.km
-1
) 1.38 1.51 1.41 0.13 7.65 0.55 0.03 1.47 0.39 0.10 -5.17 0.71
HC (g.km
-1
) 0.68 0.72 0.69 0.04 4.42 0.65 0.01 0.72 0.43 0.03 -3.25 0.71
NOx (g.km
-1
) 0.25 0.25 0.25 0.01 1.98 0.57 -0.0003 -0.37 0.60 0.01 -2.16 0.64
Fuel economy
(km.litres-1
)
6.41 6.27 6.38 -0.14 -2.46 0.66 -0.03 -0.74 0.57 -0.11 1.86 0.69
Counter-intuitively Advisory seems to have larger impact
14. Raw Data Analysis and Modeling
Data modeling [Analyte]i = te(speedi-n, acceli-n) + … +
te(speedi-m, acceli-m)
Results
ISA Mode Comparison
OFF vs. ADV OFF vs. VOL ADV vs. VOL
OFF ADV VOL difference % p difference % p difference % p
Diesel Mondeo CO2 (g.km
-1
) 258.19 266.57 260.65 8.38 3.04 0.52 2.46 0.93 0.63 5.92 -1.89 0.63
All Routes CO (g.km
-1
) 0.12 0.12 0.12 0.01 4.00 0.61 0.001 1.41 0.49 0.004 -2.34 0.82
HC (g.km-1
) 0.27 0.28 0.27 0.01 3.15 0.70 0.001 0.31 0.59 0.01 -2.61 0.82
NOx (g.km-1
) 0.74 0.77 0.75 0.03 3.40 0.46 0.01 1.05 0.60 0.02 -2.11 0.64
Fuel economy
(km.litres
-1
)
7.75 7.68 7.73 -0.07 -0.14 0.52 -0.02 -0.63 0.54 -0.06 0.84 0.56
Petrol Mondeo CO2 (g.km
-1
) 370.07 388.75 373.98 18.7 4.36 0.66 3.91 0.77 0.57 14.8 -3.15 0.69
All Routes CO (g.km
-1
) 1.38 1.51 1.41 0.13 7.65 0.55 0.03 1.47 0.39 0.10 -5.17 0.71
HC (g.km
-1
) 0.68 0.72 0.69 0.04 4.42 0.65 0.01 0.72 0.43 0.03 -3.25 0.71
NOx (g.km
-1
) 0.25 0.25 0.25 0.01 1.98 0.57 -0.0003 -0.37 0.60 0.01 -2.16 0.64
Fuel economy
(km.litres-1
)
6.41 6.27 6.38 -0.14 -2.46 0.66 -0.03 -0.74 0.57 -0.11 1.86 0.69
While more consistent, these are still not statistically significant
19. Micro-trip Analysis
Chopping total journey into a series of segments or sub-journeys
and analyzing these
So, working at resolutions
between the total journey and raw data levels
BUT most importantly
we are retaining ‘near neighbour’ information
The approach has the potential to provide a trade-off between the
two extremes of conventional analysis
20. Micro-trip Analysis
Micro-trips has traditionally been more commonly used in transport
modeling than transport monitoring
Relatively few examples from monitoring work
Example: DeFries and colleagues used micro-trip separation by
vehicle movement start/stop time, so segments were vehicle
movements steps
BUT work elsewhere, e.g. the use of rolling window averages based
of CO2 emissions in EU studies, suggested (to me at least) other
segmentation strategies could be worth considering
Reference: James E. Warila, Edward Glover, Timothy H. DeFries, Sandeep Kishan. Load
Factors, Emission Factors, Duty Cycles, and Activity of Diesel Nonroad Vehicles. 23rd
CRC Real World Emissions Workshop, San Diego, California, April 7-10, 2013.
21. Other Micro-trip Separations
Examples
• By Location
(and by extension by link, road feature, type, geometry or
conditions, etc)
• By Vehicle Activity
•By speed, acceleration, VSP event, etc
However, the associated data handling is
potentially highly time-consuming
22. This is one series of micro-trips (Marylebone Flyover, UK)
Here, we are looking at CO2 emissions (%change ISA OFF to Voluntary)
• An orange micro-trip means there is an emission penalty
• A blue micro-trip means there is an emission saving
• A red box around the micro-trip means it is statistically significant
23. Most places look like these:
• Most often a small change
• Most often a penalty rather than a saving
• Most often NOT statistically significant
24. But this stretch of road is different:
• Huge emission saving (30-70%)
• Statistically significant
25.
26. ‘Misassignment’ of speed limit means the
ISA managed vehicle is held at 30 mph
on the uphill while other vehicles
accelerate up hill to 40 mph…
So, the saving is a function of local
geography and speed limiting…
29. Sources:
Rowlingson, B. and Diggle, P. (1993)
Computers and Geosciences, 19, 627-655.
Bivand, R. and Gebhardt, A. (2000) Journal
of Geographical Systems, 2, 307-317.
Define an irregular
Polygon…
… and extract all
journey data
within it
33. …And then automate it so we can ‘daisy chain’
it for multiple micro-trips on multiple runs
34. …BUT, once you have a step like this automated,
you very quickly find extra uses for it
Three clicks: one at the center of
the target roundabout, and one
each at typical entry and exit
points, then assume circular
areas/known radii
Here, because we want
a standard area about
each roundabout, we
use a simple point and
click method to make
reference files
Here, we used
Google Maps to
measure roundabout
turning angles