"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack
Developing a simulation Model of Solid Waste Collection Operations using GPS data
1. Development of a simulation
model of solid waste collection
operations using GPS data
Dr. Bruce G. Wilson, P.Eng.
Thuy Nguyen
Department of Civil Engineering
University of New Brunswick
2. Overview
Some history
Outline the purpose of the model
Sources of data
Data analysis
Results
Other things we observed
3. History of Waste Collection Models
First computer simulation of MSW
collection published in 1965
Quon, Tanaka, and Charnes
At least 12 more since then
First paper on MSW collection route
optimization published in 1962
Mei-Ko (“Chinese Postman Problem”)
Several hundred papers since then
Very little actual progress
4. History (this project)
City of Hamilton installed GPS units on five
collection vehicles in 2003
City provided the data to the University of
New Brunswick (UNB)
UNB has produced several studies
• Potential uses of GPS data
• Queuing at transfer stations
• Estimating fuel consumption on route
• Estimating GHG emissions on route
• Simulation model in a GIS environment
5. Simulation Model - Goal
• To develop a reliable model of an MSW
collection system in GIS environment using
GPS data
• Purpose of the model is to predict:
• Time required to complete various stages of a
route
(mean and variance)
• Fuel consumption for various stages
(mean and variance)
6. Data Used
GPS Data
• 700+ days from several vehicles
• 1 minute or 15 second sampling
GIS Data
• Road network (street length & topology)
• Number of houses on each block face
Fuel consumption data
• Daily fuel records
• published rates for idling vehicles
Weigh scale records
• time & weight
7. GPS Data
GPS data provide a
snapshot at a specific
time
• position (longitude and
latitude)
• speed
• heading angle
• 1 minute or 15 sec
intervals
8. GPS Data Analysis
Need to:
• Assemble the data into a contiguous
collection route (i.e. connect the dots)
• Follow the road network
• Assign a point to a particular side of the
street (based on direction, other points)
• Infer activity (travelling, loading,
deadheading, etc.)
11. Tracing Tool Results
Actual route followed
Total distance traveled
Number and type of turns made
(right, left, & U turns)
We still don’t know what the driver
was doing at any point
13. Activity inside a Buffer Area
Each GPS point = 1 min (or 15 s)
If speed ~0, then loading waste
If speed >0
• If truck is loading waste
stop to stop travel
• If truck is not loading waste
Deadheading
Estimate average speed, average
loading time
14. Geofences
Similar to buffers, only larger
Count time spent in a specific area
Examples
• Garage
• Assigned collection area
• Unloading Facility
Transfer station, MRF
Allows calculation of travel time
between areas, etc.
16. Putting the pieces together
Simulation model built in a GIS
environment (ArcGIS 9.2, VBA scripts)
Information from 3 digital map layers:
• (i) street segments,
• (ii) collection district parcels, and
• (iii) transfer stations (TS)
Model parameters from GPS, fuel
consumption, and weigh scale records
17. 3 required map Main tool
layers to start the
model
1 collection area
(land parcel)
21. Model Validation
Model validated for 10 routes
• 5 low density, 3 medium density, 2 high density
• Compared mean and variance of time to complete
collection route
Excellent agreement on low density routes
(mean 5%, variance ~ equal)
Mixed success on medium and high
density routes (mean 15%, variance ~
equal
• Traffic?, multifamily?, street topology?
22. Other Observations
In building the model, we observed
several interesting things:
• The route followed did not significantly
impact total time on the route
• Time and fuel required to collect a kg of
waste varied significantly depending on
the route
23. 500
Drivers rarely
480
460
440 followed the same
Time (min)
route from week to
420
400
380
y = 129.25x + 266.84
week
360 2
R = 0.0584
340
320
Drivers were
300 capable of finding a
0.95 1 1.05 1.1 1.15 1.2
Collection efficiency
1.25 1.3 1.35
minimum distance
path through the
240 collection area
230
Even when they
220
210
200
followed a very poor
Time (min)
190
180
170 path, it added very
160
150
y = 47.977x + 124.75 little time to their
collection day
2
140 R = 0.0586
130
120
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 1.5 1.55
Collection efficiency
24. Collection Efficiency
Population Truck Average waste L fuel per kgCO2E per
density quantity Tonne-waste Tonne-waste
(T/day)
Low density Co-collection truck 6.56 33.5 3.5 92.4 9.6
Garbage packer 8.39 14.6 1.4 40.3 3.9
Med. density Co-collection truck 5.90 10.1 0.7 27.8 2.0
High density Co-collection truck 6.33 5.8 0.8 16.0 2.1
Garbage packer 14.93 3.2 0.4 8.9 1.1
25. Conclusions
The large quantity of GPS data allowed us to develop a
statistically reliable simulation model of MSW collection
operations
Accuracy of the model could be improved by more frequent
sampling of GPS data (~ 1 - 5 seconds)
Integrating the model with a GIS allows the model to be
interpreted visually
Many of the variations (time, fuel, efficiency) appear to be
related to housing density
Routing of trucks within a collection area may not be as
important as we’ve been led to believe
Modeling allows us to determine basic efficiency measures
26. Acknowledgements
City of Hamilton, Dept. of Public Works
Natural Sciences and Engineering
Research Council of Canada
(Dr.) Thuy Nguyen, (P.Eng.)
Betsy Agar, P.Eng.
Julie Vincent, P.Eng.
27. Questions?
Dr. Bruce G. Wilson, P.Eng.
Dept. of Civil Engineering
University of New Brunswick
Fredericton, N.B., Canada
wilsonbg@unb.ca