Content uploaded by George Asimakopoulos
Author content
All content in this area was uploaded by George Asimakopoulos on Sep 11, 2020
Content may be subject to copyright.
Dynacargo: The evaluation results of a Dynamic Waste
Collection Management System based on Real-time and
Forecasted Data
George Asimakopoulos, Sotiris Christodoulou, Panayiotis Alefragis, Andreas Gizas,
Vassilios Triantafillou
Computer & Inform. Engineering Dept.
Technical Educational Inst. of Western Greece
Antirion, Greece
{gasimakop@gmail.com, sxristod@teimes.gr, alef@teimes.gr, gizas@ceid.upatras.gr, triantaf@teimes.gr}
ABSTRACT
Dynacargo research project aims at developing a near real-time
monitoring system that monitors and transmits waste bins’ fill
level, which are used to predict the fill-level of waste bins and
dynamically manage the waste collection process by introducing
truck distance minimization, relying on efficient routing
algorithms. Dynacargo places a set of durable, low cost sensors
and RFID tags on waste bins. These tags store the fill-level
estimated by the sensors, which is passed through diverse
communication channels and ends to a central cargo information
management system. Along with this real time data harvesting,
data mining techniques are utilized on historical data collected
prior to Dynagargo implementation, in order to predict future
waste bins fill rates. In this paper we present and discuss the
promising evaluation results of the project on the case study of
Nafpaktia municipality.
Categories and Subject Descriptors
C.0 GENERAL: System architectures, H.2.8 Database
Applications: Data Mining, G.2.2 Graph Theory: Graph
algorithms
General Terms
Algorithms, Design, Experimentation.
Keywords
Urban solid waste collection; intelligent transportation systems;
data mining; graph routing algorithms.
1. INTRODUCTION
Nafpaktia municipality is a typical case of solid waste
management throughout Hellas, as waste collection is based on
standard time intervals and according to fixed vehicle routes.
Decision making is solely empirical, which leads to biased
decisions that do not take into account real needs based on data.
This has led to results far from optimal; as it is not uncommon to
have overfilled waste bins were uncollected for some days, while
at the same time other unfilled bins were collected, resulting to an
unsatisfied local society along with increased cost. Regarding
Athens[8], it has been estimated that the 60% to 80% of the total
cost of waste collection, transportation and disposal is spent
during collection. The main cost reduction efforts should deal
with distance and duration minimization of vehicle routes [6].
Johansson [7] proved that if the fill level of bins was taken into
account and waste collection adapts accordingly, it could reduce
the cost of waste collection up to 20%. Under the light of these
findings, Dynacargo (Dynamic Cargo Routing on-the-Go) project
built a near real-time monitoring system that monitors and
transmits waste bins’ fill level, in order to make waste collection
more efficient by cost reduction which is accomplished by
minimizing distances covered by refuse vehicles. In this paper we
present and discuss the promising evaluation results of the project.
2. RELATED WORK & INNOVATION OF
DYNACARGO
There are various approaches available in order to harvest data
from points of waste collection [2,3,4, 5,7,9,10]. Although they
introduce interesting approaches, they cannot be used as-is in
Dynacargo. This is because they introduce major costs, they do
not meet Dynacargo functional requirements and are mostly
customized for recyclable waste which can be managed with
longer collection cycles compared to domestic waste.
This implies that Dynacargo, despite aiming at waste collection
automation as existing approaches do, displays major
differentiations against these approaches. At the heart of these
differentiations lies Dynacargo architecture which originates from
a generic multipurpose cargo-based dynamic vehicle routing
approach which copes with cargo changes and collection from
disperse points of concentration.
Dynacargo moves forward from sole data collection from waste
concentration points, as it utilizes such data in real time in order
to optimize vehicle routes during the waste collection process
execution.
Another point at which Dynacargo differentiates from existing
approaches is the utilization of a diverse set of data transmission
techniques that is incorporated to establish data transmission from
points of collection to the system in near real time. Besides GSM
which induces fixed telecom costs, Delay-Tolerant Networking
concepts are adapted for data transmission from the disperse
concentration points to the central information system. Dynacargo
DYNACARGO Central System
Refuse Truck Subsystem (RTS)
Bins Data Gathering
and Dispatching Unit
Internet
Reader
GPS
Storage Space
Spatial DB with
the route
DTN Communication
Controller
Data Collection and
Integration Subsystem
DYNACARGO
Archival DB
DYNACARGO
Central DB
update
Routing Optimization
Subsystem
Optimized Routes
Web Portal
Data Mining Subsystem
Bin Subsystem
Sensors to estimate the
fill level of bin
data
data
data
RFID Tag
Bin’s Data and
timestamp
Navigation
Application
with GPS
Route
Modification
Application
Bin’s Data and
timestamp
Bins Data Gathering Subsystem
(BDGS)
Data
Transmission
Bins Data Gathering
and Dispatching Unit
Reader
Storage Space
Bin’s Data and
timestamp
Bin’s Data and
timestamp
Data
Transmission
Handling
Application
Data
Collector
Data
Collector
Truck
Driver
Truck
Driver
Teleconference
Application
Fleet Management
Control Center
Teleconference
Application
Map
Application
Spatial DB
Map Server
3Rd party
Geospati
al Data
WMS
Data Warehouse
Citizen
Citizen
Smartphone
App
System
Administrators
System
Administrators
FMS Server
DTN is based on a set of existing public commuters that are
utilized as data hosts that carry data as they execute unaltered
standard procedures.
3. DYNACARGO PROJECT OVERVIEW
Dynacargo functional requirements evolved from the analysis of
the solid waste collection and management system of Municipality
of Nafpaktia, Greece, which expands in an area of 870,38 km2 and
displays a population of 27.800 citizens. Nafpaktia displays
geographic and demographic aspects that sum to an ideal pilot
environment for Dynacargo. It is composed from a coastal city
with narrow busy streets in the historic center, coastal towns and
villages with population varying following tourist seasonality,
along with remote villages (up to 120 km. away) with a few dozen
residents located in mountainous areas. Another diversion is that
Nafpaktia displays waste transshipment from small to big refuse
vehicles. Requirement analysis initiated with an as-is analysis in
order to formally document the current waste management
process. Several indicative waste collecting routes were selected
and modeled on Google Earth (Figure 1), after an assessment of
peculiarities each route presents. The analysis included available
waste collection historical data and information regarding related
processes.
Figure 1. Indicative waste collection routes for Nafpaktia.
In order to reassure the accurate Dynacargo waste level
predictions, waste bin fill level data is acquired at short time
intervals from as many as possible waste collection points. Data
collection is realized through diverse data harvesting tools take
advantage of existing transportation systems such as public bus
services, postal office vehicles, taxis, municipal police vehicles,
along with active social entities as these vehicles traverse the area
of interest in an irregular but frequent manner. The Data Collector
role can be served by either a vehicle driver (e.g., bus, taxi,
postman car, etc.) equipped with the Dynacargo equipment, or
anyone else who may be involved in data collection (e.g.,
postmen) who may use any means of transportation such a car, a
bike, or by foot. Citizens improve data harvesting by reporting
fill-level with a mobile application, reporting estimations of
produced waste volume on unforeseen events and by checking
online the bin fill levels near their residence so as to discard their
waste on nearby unfilled bins.
The functional requirements and the Dynacargo architecture are
described in detail in our previous research works [1,14].
Dynacargo architecture is designed as a generic cargo-based
routing system along with the inherited adaptability to any other
municipality regardless of specific waste collecting process
characteristics. The Dynacargo architecture is presented in figure
2 and in more detail in [1]. Following, we shortly describe its
main subsystems and architectural parts.
Bin Subsystem: Sensors estimate waste bin fill level and collect,
store and transmit relevant data. The proposed smart waste-bin
system is envisaged to be mounted on the top lid of a waste-bin
and it consists of the sensing units, an active RFID tag for data
aggregation and transmission, as well as a protective enclosure for
the sensors and the RFID tag that may optionally include an
external battery source. All tags are configured to operate in
beacon mode, broadcasting the information every few seconds in
a range of about 100 meters. Active RFID tag transmission
technology is much lighter in complexity and in coping with harsh
environments compared to that of WSNs making the architecture
robust and resistant.
Bins Data Gathering and Dispatching Unit (Collector Unit):
This unit consists of an active RFID reader and a small-form-
factor computer (currently a Raspberry Pi), communicates with
the Bin Subsystem and routes the collected information to the
Central System, through a long range network. It is built of three
components:
“Reader”, which refers to the communication interface with
the bin subsystem.
“Storage space” which temporarily stores the data until
transferred to the Central System.
“Data transmission” to the Central System, through Delay
Tolerant Networking (DTN) technology.
Figure 2. Dynacargo Architecture.
The Unit could be held by any approved personnel or installed in
a vehicle of existing organized transportation systems such as
public bus services, postal office vehicles, taxis, municipal police
vehicles, etc. in order to unobtrusively collect the required
information. As the vehicle roams around the city, the RFID
reader reads the tags and collects and stores the information to the
storage area. Given the short “packet” size (less than 128 bits), the
tag range (100 meters), and a realistic average speed of 40 km/h in
a city, the tag beacon interval is extracted so as the passing
vehicle is able to complete all necessary transactions with the
smart-bin. The use of RFID technology allows reading multiple
tags simultaneously (group of nearby bins), without collisions and
re-transmissions. Communication: In order to cope with
increased telecommunication costs and infrastructure upgrades,
this Unit defers transmissions to the Central System until an
Internet connection becomes available. The Delay Tolerant
Networking (DTN) paradigm ensures that the information is
retained in the bins, until a Collector Unit passes nearby and then
in the Collector Units, until an Internet connection is available.
Bins Data Gathering Subsystem (BDGS): The BDGS is the
subsystem carried by the Data Collector. It consists of the Bins
Data Gathering and Dispatching Unit, combined with a very
simple application for handling it. The BDGS should be portable
and able to work with a battery.
Refuse Truck Subsystem (RTS): This subsystem will be
installed on each refuse truck. The main part of the RTS is the
Bins Data Gathering and Dispatching Unit. Additionally the RTS
will be equipped with a GPS unit and a camera and will
incorporate the appropriate software to provide the following
functionalities:
GPS Navigation Application: It is a classical navigation
application through GPS, which will provide navigation
guidance to the truck driver, and instructions about which
bins should be collected.
Teleconference Application: Teleconference communication
of driver with the Central System, especially for reporting
emergency situations.
Citizen Application: An application for mobile devices
(smartphone, tablet), through which the citizen can choose any
bin and report data for it (fill level, photo, comment, etc.).
Central System: It is the back-end system of Dynacargo. Its main
part is the Data Warehouse storing all historic data, bins’ data,
vehicles’ data and everything needed for calculating the best
routes based on the current load of bins and some restrictions (Bin
Collection Settings) specified by the system administrators.
Moreover, Data Warehouse will store any information derived
from the other subsystems, particularly from the Data Mining
Subsystem and the Routing Optimization Subsystem.
Apart the above databases, the Central System will include:
Fleet Management Control Center: receives the optimized
routes from the Routing Optimization Subsystem, two hours
before the start of the routes and on the fly (if the routes have to
be modified), and forwards them to the appropriate RTS. All
spatial information for the corresponding route will be stored
locally in a Spatial RTS Database. The Fleet Management Control
Center can send route modifications to any refuse truck, informing
properly the Navigation Application, if such need arises and
provided that the refuse truck has network coverage.
The Map Application: A cartographic JavaScript API that
provides all the required functionality for creating rich-web
applications based on geographic and descriptive data.
Data Collection and Integration Subsystem: This subsystem
collects bins data from the RTS, the BDGS and citizens’
applications, and integrate them into the central DBs.
Furthermore, it is able to integrate data from other sources, if
needed, like an independent FMS server that can operate in
parallel with the Dynacargo system. The architecture design of
Dynacargo does not require the existence of a classical FMS.
Data Mining Subsystem: The data mining and prediction
analysis subsystem is based on the data warehouse infrastructure,
from which the data are extracted, so that the various data mining
scenarios can be realized. It is mainly used to estimate fullness of
bins when we do not have the available information updated or is
not fairly recent. The data mining process can be quite
challenging and time consuming since the efficient
parametrization of the chosen variables requires experience on
behalf of the scientists and is mostly empirical. The research team
tried to execute different scenarios in order to analyze the
behavior of existing data mining methods on this dataset and to
investigate the existence of association rules and possibly to make
interesting conclusions. The whole subsystem is realized in a
Microsoft SQL Server RDBMS, utilizing the relevant Analysis
Server (Microsoft Analysis Services), where a Mining project has
been created in the supported programming environment, linking
the central database with the Analysis Server.
Routing Optimization Subsystem: The role of this subsystem is
the dynamic route planning before the starting of truck routes, and
the on-the-fly modification of routes, either due to exceptional
events (accident), or if the new data collected during the routes
impose such changes. The dynamic routes are calculated
whenever needed, taking into account various data that will exist
in the central data warehouse system.
Most non-urban municipalities, especially in Greece due to the
land structure, include geographically scattered villages or towns,
thus in our model we group the bins of a village or a suburb into
bin clusters. This helps the problem solution as it is usually more
economically viable to collect all bins in a village than returning
the following day to partially collect them. We defined generic
region types, i.e. Historic Center, City, Suburb, Town, Village –
Rural area, Remote area, that can be applied to any kind of
municipality.
The dynacargo routing problem can be categorized as a
Generalized Vehicle Routing Problem. Out algorithms support
several starting and ending truck points and the transshipment of
collected waste from small vehicles to larger ones. The basic idea
is to decide which bins should be collected, based on their priority
(fullness, days passed since last collection) and then calculate the
best routes (less km, less fuel, minimum cost per kilogram of
collected waste) to maximize the collected garbage given that
each available vehicle has a maximum capacity.
In order to estimate the quality of the generated solution we also
solve a sequence of instances for the next 7 days using forecasted
data. We generate solutions and evaluate if the average cost per
collected waste for the sequence horizon increases. If this
happens, we adjust the search parameters as this is an indication
that the currently generated solution leaves waste to the
uncollected bins that will create demand in the following days that
will increase the total cost.
Dynamic re-routing: When a vehicle is unable to continue its
route, we then resolve the problem using the partial operated
routes and vehicle state as fixed partial solutions. More
specifically we use: The current positions of all other vehicles,
their current available capacities, the current routes that they have
planned to operate and forecasts of bins that have not yet been
collected, now taking into account only high priority bins.
For the studied problem, the actual distance between two
geographic points is calculated by using the QGIS tool and the
Open Street Map platform via the online routing API that is
provided. Our internal representation is based on the open
standards WGS84 coordinate system, enabling us to easily extend
it to use the routing API of Google or Bing Maps in the future.
We used OR-tools [13] as the underline solver infrustrucutre,
more specifically the routing constraint solver. The heuristic
approach is implemented in Python and uses a callable CVRP
model that can use multiple depots for each subproblem. The
CTOP approach is implemented in Java and directly models the
problem as a constraint programming problem.
Refer to [15] for more detailed description of the routing
algorithms, the research issues that have emerged during the
design and development of them, and the evaluation results.
Citizens Web Portal: A web site, through which citizens can be
informed about the current completeness of bins or report bins’
data (fill level, photography, etc.) in the Central System, by
selecting a bin on a map. The purpose of the Citizen Web Portal
and the smartphone app are to motivate people to participate in
sustainable waste management.
In the next section we present and discuss the evaluation results of
the project.
4. Evaluation Results
At first routing data was calculated (number, distance, fuel
consumption, garbage load) for the current garbage collection
process for the selected routes. Afterwards, in order to assess
algorithm efficiency, Dynacargo operation was simulated for one
year (2013). We built software to produce virtual loads per bin
based on real data, like the total load weight per route throughout
the year and the empirical estimations of drivers per bin. The
result of this procedure was a set of predictions regarding bin fill
levels for 10 days of the year (1st and 15th day for some summer
and winter months of 2013) which were produced by the
prediction subsystem.
Afterwards the routing algorithm was executed for these 10 days
and number of routes, kilometers and fuel was calculated for one
week forward. The level of vehicle fullness was calculated by
dividing the actual weight of collected garbage during each week
by the total vehicle load capability. These results are displayed in
the Tables 1 & 2 for winter and summer, compared with the
corresponding weekly values of current fixed routes.
The following improvements were detected:
WINTER:
Routes: 13 instead of 21 (38.10% drop)
Kilometers: 946.6 instead of 1289.7 (26.60% drop)
Fuel: 275.4 instead of 528.16 (47.86% drop)
Vehicle load: 86% instead of 53%.
SUMMER:
Routes: 16 instead of 21 (23.81% drop)
Kilometers: 1050.4 instead of 1289.7 (18.55% drop)
Fuel: 314.6 instead of 528.16 (40.43% drop)
Vehicle load: 85% instead of 67%.
Table 1. Improvements over fixed routes for 1 week (winter)
Routes km fuel (lt)
Vehicle
fullness
Fixed Routes 21 1289.7 528.16 53%
13 949 290 81%
38.10% 26.42% 45.09%
13 938 262 82%
38.10% 27.27% 50.39%
14 974 287 90%
33.33% 24.48% 45.66%
12 914 267 90%
42.86% 29.13% 49.45%
13 958 271 87%
38.10% 25.72% 48.69%
Average 13 946.6 275.4 86%
Avg. reduction 38.10% 26.60% 47.86%
std dev 3.37% 1.74% 2.35%
Dynacargo Routes
15/12/2013
1/3/2013
1/10/2013
15/2/2013
15/11/2013
Table 2. Improvements over fixed routes for 1 week (summer)
Routes km fuel(lt)
Vehicle
fullness
Fixed Routes 21 1289.7 528.16 67%
16 1035 313 81%
23.81% 19.75% 40.74%
16 1032 306 85%
23.81% 19.98% 42.06%
16 1089 316 89%
23.81% 15.56% 40.17%
16 1049 314 84%
23.81% 18.66% 40.55%
16 1047 324 85%
23.81% 18.82% 38.65%
Average 16 1050.4 314.6 85%
Avg. reduction 23.81% 18.55% 40.43%
std dev 0.00% 1.77% 1.22%
1/6/2013
15/6/2013
1/7/2013
15/7/2013
1/9/2013
Dynacargo Routes
As Dynacargo can reach a 100% fill level per vehicle instead of
85%, a 15% margin regarding vehicle load was imposed in order
to compensate with potential forecast errors regarding bin fill
level. An improved vehicle fill level is accomplished because the
algorithm selects a vehicle with the required capacity as the load
demand of each route is known before the route execution begins.
These results when projected to an annual level yield the figures
showed in table 3.
Table 3. Dynacargo overall improvements over current
solution
Routes
km
fuel (lt)
Currently
1092
67000
27500
Dynacargo
754
52000
15400
Reduction
31%
22%
44%
For the specific case interesting conclusions can be derived:
For remote routes which display small loads there is no
significant improvement. This happens because a minimum
of 1 route per week is executed regardless of load due to the
nature of the load (waste cannot left at the bin for more days
than a specific threshold).
For remote routes which display larger loads, 1 to 4 routes
can be eliminated which is a significant gain.
The more significant improvements were displayed by bins
in the city areas of Nafpaktos area, which was not expected at
this scale. That means, that dynacargo solution can provide
significant economic profit for municipalities with cities with
population over 20000 citizens.
The specific municipality has more routes which were not
examined during the project which it is expected to display similar
improvements when analyzed. Thus we can securely conclude that
for all routes in Nafpaktia the improved figures would be:
26000 less kilometers
24000 less liters of fuel
According to accounting conducted by the Municipality of
Nafpaktia, each kilometer costs 1.8 €. Thus Dynacargo introduced
a total saving of 26000*1.8= 46800€ per year.
5. CONCLUSIONS
Sustainable growth, in regards to urban areas, requires intelligent
waste collection management. Dynacargo serves this need by
developing a cargo-centric waste transport management system
and by implementing a fully functional instance regarding
domestic waste collection in a real life large scale scenario.
This is achieved by expanding traditional fleet management
functionality in two manners. One breakthrough that Dynacargo
utilizes is to stream near real-time waste related information (fill
level of waste bins) into the monitoring and decision support
process, prior to collection from waste concentration points. If
this information is not available or not efficient from a cost-
benefit perspective, it is substituted by historical data in order to
predict waste bins status. The second breakthrough that
Dynacargo introduces is active citizen involvement, by turning
them into active information producers and consumers.
Dynacargo utilizes low-cost durable RFID tags, along with
alternative network protocols such as DTN in order to minimize
telecommunication and hardware costs.
But the most important cost reduction is accomplished by
minimizing distances covered by refuse trucks. In order to achieve
this we have introduced dedicated dynamic routing algorithms
specifically created for this kind of problems. In this paper we
briefly present the modules of the Dynacargo architecture.
Furthermore, we evaluated the project against the currently used
fixed routes in Municipality of Nafpaktia and the results showed
that the improvement could be up to 22% less kilometers and 44%
less liters of fuel per year.
6. ACKNOWLEDGMENTS
This work has been financially supported by the Greek General
Secretariat of Research and Technology and European Union
under the project “11SYN_10_456: Dyanacargo”.
7. REFERENCES
[1] Asimakopoulos G., Christodoulou S., Gizas A., Triantafillou
V., Tzimas G. and Vasilopoulos N. (2014). Functional
specifications and architecture of a dynamic routing system
for urban waste collection using real-time information.
IADIS 13th International Conference on WWW/Internet, 25-
27 Oct 2014, Porto, Portugal, pp. 203-210.
[2] Al Mamun M.A., Hannan M.A., Hussain A., Basri, H.
(2013). Wireless Sensor Network Prototype for Solid Waste
Bin Monitoring with Energy Efficient Sensing Algorithm.
IEEE 16th International Conference on Computational
Science and Engineering.
[3] Arebey M., Hannan M A, Basri H., Begum R A, Huda A.
(2010). Solid Waste Monitoring System Integration based on
RFID, GPS and Camera. Proceedings of IEEE Intelligent
and Advanced Systems (ICIAS)
[4] Faccio M, Persona A and Zanin G. (2011). Waste collection
multi objective model with real time traceability data. In
Waste Management, Elsevier, Vol. 12, No. 2,pp 391-405.
[5] Gomes T., Brito N., Mendes J., Cabral J. and Tavares A.
(2012). WECO: A wireless platform for monitoring recycling
point spots. Proceedings of 16th IEEE Mediterranean
Electrotechnical Conference (MELECON).
[6] Gukhool O., Bokhoree C., Mohee R. and Jamnejad G.
(2008). Solid Waste Collection Optimization using a Route
Time Approach: A case study for an urban area of Mauritius,
23rd Int. Conf. on Solid Waste Technology and
Management, Philadelphia PA, p501.
[7] Johansson Ola (2006). The effect of dynamic scheduling and
routing in a solid waste management system. In Waste
Management, Vol. 26, No. 8,pp 875 – 885.
[8] Municipality of Athens. (2003) Estimation, Evaluation and
Planning Of Actions for Municipal Solid Waste Services
During Olympic Games 2004. Municipality of Athens.
[9] Nuortio T, Kytöjoki J, Niska H and Bräysy O (2006).
Improved route planning and scheduling of waste collection
and transport. In Expert Sys Appl, Vol. 30, No. 2,p 223-232.
[10] Roshan I. and Akshai M. (2013). SVASTHA: An Effective
Solid Waste Management System For Thiruvalla
Municipality in Android OS. Proc. of IEEE Global
Humanitarian Technology Conf. South Asia Satellite.
[11] Toth P. and Vigo D. (2001). The vehicle routing problem.
Society for Industrial and Applied Mathematics,
Philadelphia, PA, USA.
[12] Vansteenwegen P., Souffriau W., Oudheusden D. V. (2011).
The orienteering problem: A survey. European Journal of
Operational Research, Vol. 209, Issue 1, pp. 1-10.
[13] OR-TOOLS Library: https://github.com/google/or-tools
[14] Asimakopoulos G., Christodoulou S., Gizas A., Triantafillou
V., Tzimas G., Gialelis J., Voyiatzis A., Karadimas D.,
Papalambrou A. Architecture and Implementation Issues,
Towards a Dynamic Waste Collection Management System.
Proceedings of the 24th International Conference on World
Wide Web Companion (WWW2015), pp 1383-1388, May
2015, Florence, Italy.
[15] Christodoulou S. P., Alefragis P., Gizas A., Asimakopoulos
G., Triantafillou V. Dynacargo routing subsystem and its
algorithms for efficient urban waste collection. 11th
International Conference on the Practice and Theory of
Automated Timetabling (PATAT2016), Aug 2016, Udine,
Italy.
