See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/338790110

A SUMO Based Simulation Framework for Intelligent Traffic Management

System

Article in Transportation Research Part C Emerging Technologies · June 2020

DOI: 10.18178/jtle.8.1.1-5

CITATIONS READS

26 1,706

6 authors, including:

Shamim Akhter Nurul Ahsan

Ahsanullah University of Science & Tech East West University (Bangladesh)
109 PUBLICATIONS 517 CITATIONS 2 PUBLICATIONS 30 CITATIONS

SEE PROFILE SEE PROFILE

Shah Jafor Sadeek Quaderi Md. Abdullah Al Forhad

University of Malaya European University of Bangladesh
5 PUBLICATIONS 30 CITATIONS 4 PUBLICATIONS 26 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Shamim Akhter on 01 June 2020.

The user has requested enhancement of the downloaded file.

 Journal of Traffic and Logistics Engineering Vol. 8, No. 1, June 2020

A SUMO Based Simulation Framework for

Intelligent Traffic Management System
Shamim Akhter, Md. Nurul Ahsan, Shah Jafor Sadeek Quaderi, Md. Abdullah Al Forhad,
Sakhawat H Sumit, and Md. Rahatur Rahman
Applied Intelligent System and Information Processing Lab, Dept. of Computer Science and Engineering, International
University of Business Agriculture and Technology (IUBAT), Dhaka, Bangladesh
Email: {dakhter125, nurulahsan3, sjsquaderi11, aalforhad, sakhawathosainsumit, rahatur.rahman}@gmail.com

Abstract—Continuous increments in world population Network (ANN) was proposed and implemented in [1]-
demands transportation with essential vehicle facilities and [4]. However, decision classes were made by instinct
directly effect on road traffic volume or congestion, mostly method and thus the ITMS suffers for optimal data
in metropolitan cities, and thus it needs significant clustering strategies. Nawrin et al. [5] applied
investigation, analysis, and maintenance. In these regards,
hierarchical and partition k-means clustering algorithms
an Intelligent Traffic Management System (ITMS) with a
Deep-Neuro-Fuzzy model was proposed and implemented. on the ITMS data sets. However, calculating road weight
Dijkstra algorithm is used to select optimum path from from each cluster was a chaos overlapping scenario.
source to destination on the basis of calculated road segment Therefore, building the interoperability between
weights from Deep-Neuro-Fuzzy framework. However, classifications and clustering algorithms was necessary
Deep-Neuro-Fuzzy framework needs some comprehensive and inspired us to rethink the solution for traffic
analysis, other means some simulation or emulation, and etc, management problem based on hybridization of
to proof the efficiency and workability of the model. In this clustering and classification approaches. Thus, we
paper, we are going to explore the Deep-Neuro-Fuzzy model implemented a novel Deep-Neuro-Fuzzy classification [6]
in pragmatic style with an open-source traffic simulation
to solve the weight overlapping burdens, to remove the
model (SUMO) and helps to explore traffic-related issues
including route choice, simulate traffic light or vehicular data outliers, and to avoid the overfitting problems. Input
communication, etc in our ITMS. In addition, a new GUI is data were collected by IoT based sensors [7].
developed to control the simulation input attributes and At present, the ITMS solves the decision problem,
presents the feedbacks into the traffic flow in SUMO calculates dynamic road weights from continuous
environment. Results highlight that the proposed SUMO environmental, road and vehicle-related decision
model can realistically simulate ITMS based on the road attributes in different time domains- for every hour in
segment weights from Deep-Neuro-Fuzzy model. Different everyday in a year, and takes analytical decision on
built-in routing algorithms are also used to proof the optimal route planning. Deep-Neuro-Fuzzy [6] model is
workability of this model.
responsible for generating road weights with 98.63%
Index Terms—ITMS, SUMO, traffic simulation, dijkstra,
accuracy. However, the complete ITMS is not yet been
deep-neuro-fuzzy model, road segment weight implemented in a real testbed with real traffic scenarios.
Simulation or emulation-based implementation can
explore an opportunity to expose the model and proof the
I. BACKGROUND ANALYSIS hypothesis. On these regards, Sumo Urban Mobility
Simulation (SUMO) open-source tool is used to simulate
Vehicles are one of the major needs for transportation the ITMS.SUMO [8]-[10] is a simulation framework that
in day to day life. Due to the increment of vehicle is used for microscopic and continuous road traffic
numbers, the situation turns out to be a challenging issue simulation package designed to handle large road
to route from a source to a destination in time. The networks. It uses to develop different applications
subject becomes more concerning due to the bad road including online traffic monitoring system [11], traffic
circumstances including road construction, road accident light games [12], etc.
and other environmental causes like rainfall, temperature, A Graphical User Interface (GUI) is developed to
wind, etc. Integrating IoT based intelligent traffic control the simulation. This GUI accesses the value of the
management system can help to supply information attribute and takes input from the user. User can set
related to vehicles, traffics, road statuses, and random values to the decision attributes or our IoT tool
environmental situations, and make constructive collects attributes values directly from the environment
decisions for optimum, timely, and safely routings. Thus, and vehicles. Deep-Neuro-Fuzzy framework [6] is used
a new software-based real-time bi-directional Traffic to classify those values and to formulate a weight for a
Management System (TMS) with Artificial Neural mapped road segment. Thereafter, three (3) different
algorithms including CHWrapper, A*, and Dijkstra are
Manuscript received October 12, 2019; revised May 15, 2020. used to trace the optimum route. The result section

©2020 Journal of Traffic and Logistics Engineering 1

doi: 10.18178/jtle.8.1.1-5
 Journal of Traffic and Logistics Engineering Vol. 8, No. 1, June 2020

highlights the simulated time for all the algorithms and simulation, the simulator needs to configure the
concludes their performances. configuration file on “map.sumocfg” [14]. The
configuration follows the following format:
II. PROPOSED SIMULATION MODEL
Realization of the traffic density at a particular <configuration>
intersection for a given time can also help in reducing <input>
traffic congestion at that point. This data can be analyzed <net-filevalue="map.net.xml"/>
to determine several factors like green light length, traffic <route-filesvalue="map.rou.xml"/>
at the particular time. Our ITMS model has the capability <gui-settings-filevalue="gui-
settings.cfg"/>
to crawl data intelligently from online web domains, to
</input>
collect locations/area/road segments base high-resolution
information from IoT sensors for real time TMS decision, </configuration>
and to circulate them to the running vehicles. However,
the full model workability is not tested in real/ simulation The environmental interface of SUMO can be changed
platform. Sumo Urban Mobility Simulator is used for in “gui-setting.cfg” file and it looks like the following:
road vehicles and traffic light simulation. It uses the
concept of adaptive traffic light control algorithm and <viewsettings>
manipulates both the sequence and length of traffic lights <delayvalue="300"/>
in accordance with the detected traffic [13]. The <schemename="real world"/>
algorithm uses real-time data like the waiting time of </viewsettings>
vehicles, volume of traffic in each lane, and etc. Finding
the best and the shortest path to destination can be used as
a tool to minimize the traffic along a path. The traffic
along the road can be sent to the incoming vehicle
proving them the idea about the road traffic network and
thus they can take an alternative path to the destination
[13].
Sumo Urban Mobility Simulator is used for road
vehicles simulation. SUMO uses three (3) built-in
algorithms including Dijkstra, A*, and CHWrapper to
find out the best routing (shortest path) in the simulation Figure 1. Prototype Map (Sample Map).
environment. We apply all of them in our road map and
compare their performances. C-means clustering and
Deep-Neuro-fuzzy classification [6] tool is used to
generate road weights of a particular location. Python
code is used to generate a weight file and connect with
GUI developed by C#. Two (2) different potential state
changes are considered in our implementation-natural
occurrences (humidity, rainfall, peak hour, temperature,
and wind) and artificial occurrences (road accident, road
construction, and road status). Both types of attributes are Figure 2. Map of Gulshan, Dhaka (Real Map).
accessible simultaneously to generate a significant road
weight. The speed value is calculated from the weight TABLE I. RANGE OF ATTRIBUTES
value (proportional to weight value) Attributes Range Value Named As
A. Setup Sumo Simulation
Humidity 32-95
A certain part of Dhaka city map is downloaded from
the internet (openstreetmap.org) as .osm format. At first, Peak Hour 8-129
ITMS simulation is implemented in a prototype map
(“Fig. 1”) selected from the real map (“Fig. 2”). Rainfall 0-30
Thereafter, the whole simulation is implemented on the
real map of Dhaka city. SUMO needs to convert the map Temperature 13-34
files to XML format for generating optimum route and
Wind 0-15
traffic system. DOS command script “start-command-
line.bat” and “netconvert” command are used to convert Road Status 0,1 normal
the map.osm files to map.net.xml files. Additional
necessary elements including road name, traffic light Road Construction 0,1,2 low, medium, high
timing, etc. are added using SUMO Neteditor. The route
for vehicles is written in map.rou.xml file. “Fig. 3” shows Road Accident 0,1 no, yes
the flowchart of the full simulation. For vehicle

©2020 Journal of Traffic and Logistics Engineering 2

 Journal of Traffic and Logistics Engineering Vol. 8, No. 1, June 2020

(which is the specific road weight of number from

optimum clusters in the combined data set). A supervised
Deep Neural Network is trained to formulate the road
weight (road weight cluster number) using the
implications from the fuzzy rules. The Deep-Neuro-
Fuzzy provides a complete package for generating
dynamic road weight from given input data sets. However,
this model is sufferings for lack of proof of its work
ability and proper user interface (UI) to collect or show
I/O data. We develop our own GUI in C# to make the
simulation process interactive and user-friendly. This
GUI helps to enhance the testing process by changing
different weights to the system. Weight calculation
depends on eight (8) different attributes including
humidity, peak hour, rainfall, temperature, wind, road
status, road construction, and road accident. All of these
attributes are combined with different ranges as presented
in Table I. The selected attributes from GUI are placed to
generate the weight value by using C-means clustering
and Deep-Neuro-Fuzzy software [6]. The weight output
range is 0-10. “0” is the lowest weight value that means
the natural and road condition are in the best situation, for
that reason the roads get the maximum speed to ensuring
faster vehicle passing. “10” is highest weight value,
which means the natural or road or both are in an odd
situation, for that reasons the roads (or lanes on-road) get
the lowest speed value and vehicle passing would be
stopped or reduced their speed.

III. RESULT ANALYSIS

A. Testing GUI and Traffic Flow
We investigate the following two (2) case studies to
make changes in GUI input attributes:
CASE (1): Due to the natural phenomenon occurrences,
vehicle passing can be stopped or vehicle speed will be
decreased in the road segment(s). “Fig. 4” and “Fig. 5”
present the situation. Where changing the input attributes
in GUI, generates a new weight, calculates a new
corresponding speed, and sets automatically to that
specific road segment(s) (edge 2 and 11). Output presents
in “Fig. 5”. Although the vehicles get the green signal
and clear route, however, vehicles are not passing
through those road segments due to the high weight
Figure 3. Flow chart of full simulation process. occurred by heavy raining and huge wind (makes higher
weights). CASE (2): Similarly, any construction or
B. Formulation of Road Weights accident occurs in the road segment (edge 5) makes itself
Deep-Neuro-Fuzzy is a hybridization model of fuzzy busy. Thus, vehicle flow in this segment decreased or
C-mean clustering, fuzzification system and deep neural stopped. We apply a road construction for a road segment
networks. The compact module solves the road weight as a medium level in the road weight attributes in “Fig. 6”.
measurements problems in ITMS [1]-[3]. C-mean was This change directly reflects the simulation results
used to provide two outputs including optimum clusters presented in “Fig. 7”. Here vehicles are avoiding that
in each weather features data set and optimum clusters in road segment and find out other available free routes to
the combined data set using DI [15] and SC [16] travel to destination. Table II present few input samples
validation technique. The number of clusters in combined sets and their corresponding weight and speed values.
data sets reflects to the optimum number of road weight
B. Analysis of Different Route-Finding Algorithms
classes. Thereafter, applying the triangular membership
function and specific fuzzy inference rules on the cluster SUMO applies Dijkstra algorithm for generating
numbers those are generated by C-mean clustering optimum routes from source to destination. Three (3)
algorithms on each ITMS feathers. Implications of the more additional algorithms are interacted with ITMS and
fuzzy rules are aggregated to generate fuzzy output set try to find out the best optimal performer in the input

©2020 Journal of Traffic and Logistics Engineering 3

 Journal of Traffic and Logistics Engineering Vol. 8, No. 1, June 2020

maps. We experiment on 1,430 vehicles on SUMO TABLE II. WEIGHT AND SPEED VALUES FOR SOME SAMPLES.
simulation and apply the above algorithms. Their results Attributes

Speed Across Weight

are presented in Table III. According to Table III, the best

Calculated Weight
result comes from CHWrapper algorithm as it takes the

Road Accident
Construction
shortest path for 1,430 vehicles in optimal time-

Temperature

Road Status
Humidity

Peekhour

Rainfall
consuming. Second is considered a time unit for all

Wind

Road
algorithms. However, it is changeable according to the
CPU performance and simulation configuration.
80 15 4 21 5 0 0 0 3 30
70 40 0 28 2 0 0 0 7 15
60 78 0 24 1 0 1 0 9 8
65 10 0 20 4 0 0 1 10 5

TABLE III. ANALYSIS OF SIMULATION ALGORITHMS (TIME

CONSUMING)
Algorithm Time (seconds)
Dijkstra 3060
A* 3506
CHWrapper 2689
Figure 4. Sample input for natural occurrences.
IV. DISCUSSION
In this article we successfully implemented our ITMS
model in SUMO simulation environment and test the
model work abilities. Deep-Neuro-Fuzzy model is
working well and maintains adequate accuracy. However,
the model is not yet been implemented in real time road
networks. Real time implementation requires
personalized infrastructure for mobile devices and such
implementation is costly and time consuming. In our
Figure 5. Road state changes due to weather change
simulation we consider changing only on road weights
parameters; however, other hidden parameters of the
existing SUMO algorithms are remaining same. Thus, we
need to explore the SUMO model without build-in
algorithm so that we tune an experimental setup by
changing the algorithms parameters. The new algorithm
design in SUMO environment is a challenging task due to
its complex architecture. In addition, few scripts are
needed to (run explicitly) provide to input to SUMO
environment. Our simulation results of three (3)
algorithms highlight similar performance according to
their theoretical basements, CHWrapper performs better
than A* and Dijkstra performs slower.
Figure 6. Specially changes due road construction In near future more robust dynamic and optimum
route-finding model including ant colony or
reinforcement-based route findings algorithms will be
integrated in SUMO. IoT sensors are not yet directly
integrated with simulation ITMS model. In near future, a
personalized mobile network infrastructure with
V2Vcommunication, edge-fog computing support, and
cloud-based data storage will be implemented.

V. CONCLUSION
We successfully implemented the ITMS with open-
source traffic simulation model (SUMO) and did a
comprehensive analysis to explore traffic-related issues
including route choice, simulate traffic light or vehicular
Figure 7. Vehicle flow stopped due to construction
communication, and etc. ITMS simulation model

©2020 Journal of Traffic and Logistics Engineering 4

 Journal of Traffic and Logistics Engineering Vol. 8, No. 1, June 2020

highlights that the Deep-Neuro-Fuzzy tool is capable to [4] S. Akhter, M. R. Rahman, and A. Islam, “Neural Network (NN)
based route weight computation for bi-directional traffic
change the weight output according to the changes in the
management system,” International Journal of Applied
attributes and proportionally helps to improve the route Evolutionary Computation, vol. 7, no. 4, pp. 45-59, 2016.
situation. It also presents that the road weight is highly [5] S. Nawrin, M. R. Rahman, and S. Akhter, “Exploreing k-means
proportional to the speed. In near future, we will with internal validity indexes for data clustering in traffic
management system,” Thomson Reuters Master Journal List and
investigate real-time road attributes and create automatic
ISI index. International Journal of Advanced Computer Science
maps to show dynamic road condition to ITMS’s users. and Applications, vol. 8, no. 3, pp. 264-272, March 2017.
[6] S. H. Sumit and S. Akhter, “C-Means clustering and deep-neuro-
CONFLICT OF INTEREST fuzzy classification for road weight measurement in traffic
management system,” Soft Computing, Springer, vol. 23, no. 12,
The authors declare no conflict of interest. pp. 4329–4340, June 2019.
[7] M. A. A. Forhad, M. Nadim, M. R. Rahman, and S. Akhter, Cloud
IoT Based Mobile Agent Framework for Real-time Traffic
AUTHOR CONTRIBUTIONS Information Acquisition, Storage and Retrieval, ‘Smart Devices,
Applications, and Protocols for the IoT’, Chapter-II, IGI Global
This work is supported by Ministry of ICT division Publisher, 2019, pp. 14-32.
Bangladesh. 1st author is the principle researcher of this [8] D. Krajzewicz, J. Erdmann, M. Behrisch, and L. Bieker, “Recent
funding project and all co-authors are the research development and applications of SUMO- Simulation of urban
assistants working under his guidance. The 1st author also mobility,” International Journal on Advances in Systems and
Measurements, vol. 5, no. 3-4, pp. 128-138, 2012.
is the corresponding author of this article. This is a group [9] K. Daniel, H. Georg, F. Christian, and W. Peter, “SUMO
research work and co-authors have responsibility to (Simulation of Urban MObility); An open-source traffic
develop different parts of the whole project. Mr. Md. simulation,” in Proc. 4th Middle East Symposium on Simulation
Nurul Ahsan and Mr. Shah Jafor Sadeek Quaderi are and Modelling, 2002, pp. 183-187.
[10] F. Khodadadi, S. J. Mirabedini, and A. Harounabadi, “Improve
responsible to develop the SUMO testbed whereas rest of traffic management in the vehicular ad hoc networks by
the authors are involved in the development of Deep- combining ant colony algorithm and fuzzy system,” International
Neuro-Fuzzy tool. Journal of Advanced Computer Science and Applications, vol. 7,
no. 4, pp. 44-53, 2016.
[11] M. Krumnow and A. Kretschmer, “Real-Time simulations based
ACKNOWLEDGEMENT on live detector data – experiences of using SUMO in a traffic
management system,” Simulation of Urban Mobility Lecture Notes
The research is supported and funded by the ICT in Computer Science, vol. 8594, pp. 136-145, 2013.
ministry of Bangladesh. The authors would like to thank [12] Y. Xu, D. Li, Y. Xi, and S. Lin, “Game-Based traffic signal
to the ICT ministry of Bangladesh, International control with adaptive routing via V2I,” in Proc. 21st International
University of Business Agriculture and Technology Conference on Intelligent Transportation Systems, 2018.
[13] S. P. Biswas, P. Roy, A. Mukherjee, and N. Dey, “Intelligent
(IUBAT) and East West University (EWU) Bangladesh. traffic monitoring system,” in Proc. Second International
Authors also acknowledge to the anonymous reviewers Conference on Computer and Communication Technologies, 2015,
for their critical assessment and comments on the work. pp. 535-545.
[14] SUMO User Documentation. (n.d.). [Online]. Available:
https://sumo.dlr.de/wiki/SUMO_User_Documentation
REFERENCES [15] U. Maulik and S. Bandyopadhyay, “Performance evaluation of
[1] M. R. Rahman and S. Akhter, “Real time bi-directional traffic some clustering algorithms and validity indices,” IEEE Trans
management support system with GPS and web socket,” in Proc. Pattern Anal Mach Intell., vol. 24, no. 12, pp. 1650–1654, 2002
15th IEEE International Conference on Computer and [16] S. Aranganayagi and K. Thangavel, “Clustering categorical data
Information Technology, UK: Liverpool, 2015. using silhouette coefficient as a relocating measure,” in Proc.
[2] M. R. Rahman and S. Akhter, “Bi-directional traffic management Conference on Computational Intelligence and Multimedia
with multiple data feeds for dynamic route computation and Applications, 2007.
prediction system,” International Journal of Intelligent Computing
Research, Special Issue, vol. 7, no. 2, pp. 720-727, 2016. Copyright © 2020 by the authors. This is an open access article
[3] M. R. Rahman and S. Akhter, “Bi-directional traffic management distributed under the Creative Commons Attribution License (CC BY-
support system with decision tree based dynamic routing,” in Proc. NC-ND 4.0), which permits use, distribution and reproduction in any
10th International Conference for Internet Technology and medium, provided that the article is properly cited, the use is non-
Secured Transactions, London. United Kingdom, 2015. commercial and no modifications or adaptations are made.

©2020 Journal of Traffic and Logistics Engineering 5

View publication stats