Luận văn multi agent system for traffic simulation in vietnam hệ thống Đa tác tử Áp dụng cho vấn Đề mô phỏng giao thông Ở việt nam

List of Abbreviations MAS — Mulli agent syslem ABM Agent Based Model VIS Vietnam ‘Traffic Simulator VISSIM — Visual ‘lraffic Simulation System KDT —Khuat Duy Tien TDII-— Tran Duy IIu

DAI HOC QUOC GIA HA NỘI TRƯỜNG ĐẠI IIỌC CONG NGIIE

LUẬN VĂN THẠC SĨ KHOA HOC MAY TINH

NGƯỜI HƯỚNG DẪN KHOA HỌC: PG5 T5 BÙI THẺ DUY

TIả Nội - 2013

ORIGINALITY STATEMENT

‘T hereby declare that this submission is my own work and to the best of my

knowledge il contains no materials previously published or wrilion by another

person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at University of Engineering and

Technology (UET/Coltech) or any other cducational institution, cxcopt where

due acknowledgement is made in the thesis Any contribution made to the

rescarch by others, with whom I have worked al UET/Collech or elsewhere, is explicitly acknowledged in the thesis | also declare that the intellectual content

of this thesis is the product of my own work, except to the extent that assistance

from others in the project’s design and conception ar in style, presentation and

linguistic expression is acknowledged.’

Hanoi, October 7th, 2013

5igned che ee eeeeeee "

ABSTRACT

The VTS is a system that allows users to design different road systems as well as

to create different simulation scenarios with different agent profiles It was built

in 2010 based on the theory of Agent and Muln Agent System During 2011 and

2012, it was improved and many experiments were performed regarding to the

real daia collected from VOV traffic online The results are promising and we

hope that it could be able to help the traffic planners to solve the sore issues of traffic in Vietnam at the moment

PUBLICATION

*The Duy Bu, Duc Hai Ngo, Cong Tran Mului-agent based Simulation of

traffic in Vietnam, 13th International Conference, PRIMA, Kolkata, India, pp 636-648, 2010

1.1 Motivation, objectives and approach

1.2 Outline of the thesis

Chapter 2 Literature Review

2.1 Models of traffic simulation

Chapter 3 Vietnam Traffic Simulator

3.1 Introduction to multi-agent system

3.11 Agent

3.1.2 Multi Agent Systems —MAS

3.13 Agent based model ABMI

List of Figures

Figure 1 VISSIM visual interfaces

Figure 2, VISSIM statistic fiction

+ igue 3.Tra[lc light simulation system

igure 4, Highway simulation system,

Figure 5, A part of highway road

Figure 6, Road Area

igure 7, Connection road areas,

Figure 8 Detecting possible collisions

Figure 9 Intertace and simnlation of the traffic light

Figure 10 Some examples of real time traffic data -

Figure 1 dw Khual Duy Tien - Tran Duy Hung coward in the sitar

Figure 12 the Khnat Duy Tien Tran Duy Hung crossroad captured by the traffic camera

Figure 13 Distritmtion of inflow vehicles in real daa ceceseteteseenenaenieananatmienesass

Figure 16 The best case oŸ achieveđ zestlts

tripure I7 Nonmalization of achieVed r€SHÌS «is tàn HH ng ru

Figure 18 The decrease rate of velocily

Figure 19 Traffic light data observation

List of Tables

Table 1 Án example of randomized parametes nteeierirrrarreroooou.28

Table 2 Parameters of KDT — TDH crossroad

‘fable 3 ‘The information query form

Table 4 Án example of query data LH HH HH HH HH nga rueg

‘Table 5 Default parameters oÊthe sỉnmlation s tt nrrrarreoooui2

List of Abbreviations

MAS — Mulli agent syslem

ABM Agent Based Model

VIS Vietnam ‘Traffic Simulator

VISSIM — Visual ‘lraffic Simulation System

KDT —Khuat Duy Tien

TDII-— Tran Duy IIung

PH — Pham Hung

HL — Hoa Lae

vi

Acknowledgement

First and foremost, I would like to express my deepest gratitude to my

supervisor, Ass Prof Bui The Duy, for his patient guidance and continuous

support throughout the years He always appears when [ need help, and responds

to queries so helpfully and promptly I would like to give my honest

appreciation to my co-partner Ngo [uc Hai for his kindly support although he

had to prepare for his study oversea I would also like to thank my friend, Vu

Tien Thanh, for his kindly help I sincerely acknowledge all my locturcs in

University of Engineering and Technology, Vietnam National University, Hanoi,

for guidance in my master study Finally, this thesis would not have been

possible without the support and love of my family Thank you!

vil

Chapter 1 Introduction

1.1 Motivation, objectives and approach

In Vietnam, the transportation system is now facing many problems in terms of

congestions and accidents Especially in big cities like IIanoi, the transportation

system is chaotic, duc to narrow roads, increasing number of vehicles, and lack

of consciousness to follow the traffic rules from participants Many solutions have been proposed and implemented which imposed a great effect on the

development of the transportation system itself as well as awareness of the

whole society However, most of these solutions usually require a huge financial

cfforl to be able to prove cffeclivencss Therefore, a method which helps reduce

the cost of improving the current transportation situation should draw attention

of researchers It is the reason why I was motivated to do my thesis regarding to

this theme

In developed countries, transportation planners always have to have a strategic

vision which can identity a clear plan to develop the transport system Such

knowledge could be attamed by experimenting on traffic simulators With

information provided by Lhese simulators, the policy makers can figure a way to

reduce the cost of traffic infrastructure building Literally, the use of multi agent system in sirnmulating the behavior of the society is a common trend of solving

problems like transportation Following this wend, we starlcd Ww build the

Vietnam ‘lraffic Simulator (V'I'S) based on the multi agent system model under

the guidance of Asse Prof? Bui The Duy in 2010/17] This thosis mainly aims lo

strengthen the correctness of the VTS

To phrase il another way, the completion of the evaluation for Lhis simulator is

the main target of this thesis It requires some approaches in both proactive and

reactive ways With the base knowledge acquired from research of MASs, I

added some additional {catures and improved the structure of VTS to be more suited for the traffic in Vietnam To be more specified, | had gathered data from many sources, had added a function, had corrected the behavior of the simulator and then T implemented to evaluation phase

1.2 Outline of the thesis

“The outline of the thesis is as following: Chapter 2 will be the literature review about traffic simulation models and the approach based on the Multi-agent

model Chapter 3 is aboul some main [calures of Victnam Traffic Simulalor,

including some new improvement after the short paper presented in PRACSYS

2010 [17] The evaluation stops will be proscnicd in chapter 4 The last chapter

is the conclusion and future research.

Chapter 2 Literature Review

2.1 Models of traffic simulation

Traflic simulation can be used lo: lind treatments lor a problem of a traffic

system, test new designs of transportation facilities before the commitment of

resources to construction, analyze safely of a system; or tram traffic

management persennel [6] Due to the complexity of the transportation system,

there are two ways of modeling it with regards to scope and time In this section

we will mureduce some models that arc used lo be the base stones of Vietnam

‘Traffic Simulator with regards to three categories: time, scope and multi-agent

based system

211 Scope

Simulation models of traffic can be categorized by level of detail: macroscopic

(4, 8], microscopic [1, 9, 10, 11], mesoscopic [2, 7], and nanoscopic [3] A

macroscopic model describes entities and their aclivilies and interactions al a

low level of detail Kor example, the traffic stream may be represented in some

aggregate manner such as a statistical histogram or by scalar values of flow rate,

densily and speed A microscopic model describes both the system entilies and

their interactions at a high level of detail A mesoscopic model generally

represenls mosl entilies at a high level of delail but describes their activities and

interactions at a much lower level of detail than would a microscopic modcl

With nanoscopic models, nano simulation attempts to model drivers’ steering

behaviour and more detailed components of perception-reaction time in order to

depict the the human performance

* Visual Trafic Simulation System

In this subsection we introduce a system named Visual Traffic Simulation System (VISSIM) which is one of the mesoscopic systems developed by Thomas Fotherby [14] This system is diverse in many kinds of traffic systems simulations In details, it provides function to design the transportation

infrastructure with detailed information of flowing vehicles such as numbers of

cars, trucks and their velocities

Figure 2.VISSIM statistic fuction

This system consists four application components: Road Network Designer,

Traffic Modelling Designer, Visual Simulation, Application Results We will

summarize some main features of these components that are being used as suggestions for VTS as below:

* Road Network Designer

This section of the application should allow a user to quickly design simple

schematic road diagrams (road networks)

Variables

- The application should be to-scale (e.g x screen pixels per metre)

- The application should start by showing a drawing panel as a blank

designing arca (representing a xx m square arca)

- Assume terrain is always flat (a simplification)

Drawing lechniques

- Lanes are drawn on the panel in straight-line sections Each new

section of the lane fallows on from the previous section

- Any cxisting lane should be able to be extended with another identical

lane next to it (space permitting)

- Ensislng lanes should be able to be deleted

- Road designs should be able to be saved and loaded

« Traffic-Madelling Tool (Pre-condition: a valid road network.)

¥or the application to be realistic and produce useful results the user must be

able lo specify the traffic data that the simulalor will use This dala may be

based on real observations obtained from electronic detection devices and traffic

surveys.

Road network Lacts:

A junction has inputs and outputs

Each junction has unique input and output traffic-fow intensilies

The input traffic-mlensity of one junction will be a function of the

output traffic-intensities of other junctions

A cerlain amount of cars will enter the syslem according lo some kind

of control element

A car ean start at any input into the system and go to any output

All cars should cventually exit the system (No infinite loops)

Requirements

For each input to the system users should be able to specify the

average or exact number of cars per minute that will enter ‘his will

require labelling of the roads in the designed road network

There should be an option to randomise the car inpul dala each time

the simulation is run, or otherwise the simulation will run with

precisely the same data (the same number of cars enter at the same lime)

Panel features:

Traffic-low models should be able lo be saved and loaded

* Visual Simulation (Pre-condition: A valid road network.)

This soction should present animated graphics with drawn-lo-scale vehicles

moving through the geometry of the system ‘lhe traffic that is animated is generated and controlled according to statistics specified by the "traffic-

modelling tool"

Vehicle behaviour model:

Cars obey a speed limit This is their “top speed" An example maybe

between 50) and 60) kilometres/hour (31-37mph)

Cars enter the system at top speed at positions and times according to a

set IrafTio-model specified by the "traffic-modelling tool"

Cars do not collide.

- Cars can only onter the system if they are in a valid posilion (Not a

collision)

- Cars can only change Janes al junctions (No U-turns)

- Cars will always Wy to go at their top spced when possihle but their

speed is governed by the "car-following model" described below

- Cars can only change speed by ac

Joraling or decelerating Acceleration will be a constant value (for example Sm/s2) De-

acceleration measures should be sensible (Le a car should not be able

to stop in no timc)

- Cars do not take independent decisions A car travel route and the lane

it is in depends entirely on its starting position and the statistical

decisions of the junctions it passes through

Car following model

- A Car will travel al ils top speed limit unless it is withm 10m of

another car

- It must de-accelerate to match the other cars speed by the time there is

a 3m distance

- [t must never go within 1m of another car on the same lane

Car pull-up model

- Cars follow this model when pulling up to red lights, give-way signs or

if there is stopped traffic ahead

- Ata suitable distance before the obstruction the car will de-aceclerale

with a constant value to stop in time

Lane changing behaviour model:

- The project is simplified to not include overtaking

- A car will only change lane at junctions according to the junction

traffic-model statistics

Vehicle behaviour at give-way junctions

- Cars on the main route are unaffected and travel as normal according

lo the car-following model.

- Cars on the slip roads “pull up" to the give-way line lo check for

oncoming traffic

- Cars on the slip-road can join the main-roule if they aren't going to

obstruct the cars on the main-routs Lc there must be a suitably large

clear section of traffic on the main route This is the gap-acceptance

model Vehicle behaviour model at signalled junctions

- Signals are independent for each input lane

- Cars will "pull-up" to the stop line if the signal is red

- The signal is two-phase Go is green, stop is red

- On a green signal the car is specified an output lane (according to the

traffic-model of the junction) and will travel to the output lane in a

direct route

- Traffic light ming intervals will be initially split fairly between

different sets Later, traffic lights can be re-programmed ta he more

intelligent

- The colour of a trailic light will be conveyed on the sercơn by the

colour of the stop line at a particular lane In addition if the light for a

lane is green there should be arrows displayed on the junction

specifying where ears have the option of going

Dynamic traffic controls:

- For cach input lo the system there should be a control to increase or

decrease the traffic entering at that input

«Application Results

Each component of the simulated traffic system should log data:

- Lach input and output of the system should have a log of how many

cars passed through

- Every junction should log how many cars passed through each input

and output lane

- Each traflic-light junction should store the liming intervals of cach

light

- There should be traffic flow data for the syslem as a whole (Number

of cars passed through per sccond)

- There should be a value estimating total surface area of road surface

used in the current network design

*All of these features are implemented in the VTS

21.2 Time

‘Time is a basic dependent variable in almost all traffic simulation models

Continuous simulation models describe how the elements of a system change

slate continuously over time in response to continuous stimulation Discrete

simulation models represent real-world systems by asserting that their states

change abruptly at points in time There are generally two types of discrete

models: discrete time (e.g [9, 1]) and đisereLe ovent (e.g |7[) With discrete

time models, activities which change the states of the system elements are

compuled within each time interval The discrete event models only perform the

calculation based on the happening of events

¥* Intelligent Traffic light control system

in this subsection we introduce a simulator named Green Light District Simulator developed by Utrecht University (Netherland) [7] This is a system

which supports the determination in duration of tralTic lights

Figure 3.Traffic light simulation system

Basically, it is a microscopic traffic simulation system However, it also includes

discrete events based on time and some parameters such as the density of traffic,

average velocities of vehicles, etc These information are used to automatically

suggest the duration of traffic lights

The main components of this system are:

- Drive Lane consists of two parallel lines

- Road made by 2 Drive Lane It includes information about direction, incoming and outgoing gates which form the transportation network

- Node is the term describing cross cuts between conjuction and crossroad

- EdgeNode describes areas in which cars go in and out

- Sign describes the traffic lights These places are the points where the

duration adjustment algorithm is deployed automatically

10

- Cars play an imporlant role in the simulalion However, duc to the

main target of this simulation is automatic traffic light adjustment, the

model of moving vehicles is simplified

During the simulation phase, the system gathers statistical data such as density,

number of incoming and oulgving vehicles in order lo provide parameters for

the duration adjustment algorithm of traffic lights

2.1.3 Multi-agent system for traffic simulation

As a powerful tool of microscopic simulation, multi-agent based simulation has

been used for traffic domain, ¢ g [13, 11] Giving each vehicle three subsystems,

including Controller, Sensors and Driver model, Sukthankar et al [13] have

simulated every detailed movement of vehicles By calculating the movement of

cach agent based on finite stale machine, Wan and Tang [11] have sumulated a

traffic flow which comprises of autonomous agents/vehicles Both systems use 3D graphics to display the simulation

* Simulated Highways for Intelligent Vehicle System

This is a simulator developed by Rahul Sukthankar, Dean Pomerleau and

Charles Thorpe [13] The name of this system is Simulated Highways for

Intelligent Vehicle Algorithms (SHIVA) and it is a microscopic traffic simulation system due to the exquisite length of the highway in constrast with

the low density of trailic lights and houses Usually, the speed of the vehicles

travelling on the highway are really high, it is the reason why this system focus

on calculation of the details of the vehicles to ensure the safety of the highway

1

Figure 4 Highway simulation system The system includes 2 models: Highway road model and transportation model

© Highway road model

This model describes a net of highway road including many long roads

connected together The basic element is called RoadSegment The width of these roads are varies, but they are always equal to a multiplier of a number

called “lanewidth” Besides, they also include information about some narrowed

part called RoadSlice, the connector to connect different roads together and the maximum velocity of the vehicles travelling on that Segment

12

Road Segment Number of lanes: 3

Surface: PAVED

Figure 5 A part of highway road

* Transportation model

This is a complicated model, it includes 3 components: Controller, sensor, driver

- Controller component: this component will control the behavior of the

drivers including angle and direction steering and speed adjustment

- Sensor: This is the component which allows the driver to sens other vehicles’ behavior such as: increasing speed, decreasing speed,

steering It is flexible enough for the controller component to function

- Driver: This component is the component making decisions such as

Jane choosing Based on the information gathered from sensor

component, an algrorithm will be executed to calculate the most

intelligent decision of the drivers

The system allows users to define different types of big vehicles such as: trucks, cars, containers They can define not only the size but also other parameters

such as the ability to adjust speed

13

22 Conclusion

Recent section is about some main features of some traffic simulation systems

that we conducted research Most of these systems are microscopic system; it

means they simulate the behavior of drivers in details Furthermore, these

systems allow users lo create a flexible road system Moreover, they all have a

fully supported report component A traffic simulation system takes a “scenario”,

eg a road system or a highway network configuration, and produces the

simulation resulls in two [ormats: slalistical and graphical Quantilalive

descriptions of what is likely to happen can be provided by the statistical results

while the graphical and animated results can provide the user with insights to

understand why the system is behaving this way

Tn the next section, we will briefly introduce our model used in a simulator named V'I'S which applies the multi-agent based model

14

Chapter3 Vietnam Traffic Simulator

‘Through experiences and knowledge gained by learning these models, it is our

decision to choose a model to apply for the traffic situation in Vietnam As

stated above, the Vietnamese transportation system is more complicated than

other systems oversea due to the existence of motorbikes - the flexible and convenient but indiscipline vehicles What do we mean about this indiscipline

vehicle is that it usually goes unpredictable, for example, it can turn right or left

or even turn round back without regards to the traffic rules Follow that, cars or

buscs sometimes also make complicatcd moves In short, they do nol give way

according to the traffic rules he problem is much worse when the traffic

participants do not recognize the benefit of following rules As a result, they

behavior follows their instinct For cxample, logically wo can understand that

when waiting for the vehicles in front, stopping in the intersection is obstruct-

ing the Irafe Mow Bul instead of wailing on their lancs, the trafTie participants

will always try to fill in any space in front of them, or cven to the left or the right

of the opposite lane if there is obstruction in front In addition, the road in VicInam is also morc complex wilh many dillcront structures, narrow toads, small crossroads center, etc It is ought to understand that the simulators

presented cannot be used a tool to solve this problem since the behavior of the

entitics in these simulators arc much simpler That is the reason why we decided

to build a whole new simulator which is dedicated to be the best suited for the traffic in Vietnam

As described, the simulation of traffic in Vietnam has too many different parameters Thus, finding a mathematic model lor this problem is very difficult

‘The multi-agent based model is our choice to apply to the simulator we were

going to build at that time It has a flexible capacity which allows interactions

15

between drivers and drivers to cnvironment (road scgments, trallic lights)

‘Therefore, it is an advantage to model the traffic situation in Vietnam currently

3.1 Introduction to multi-agent system

Tn this section, we would like to introduce a few basi

There arc many concepts of agent given, but so far none of them has becn

considered as a standard concept for the agent, for example:

* “Most oflen, when poople use the torm ‘agent’ they refer to an onlily that functions continuously and autonomously in an environment in which other

processes take place and other agents exist.” (Shoham, 1993);

« “An agent is an cnlily that senses ils cnvironment and acts upon il” (Russell, 1997),

* “The lerm agent is used lo represent wo orlhogonal entilies The first is the

agent’s ability for autonomous execution ‘The second is the agent’s ability to

perform domain oriented reasoning.” (the MuBot Agent);

* “Intelligent agents are software entities thal carry oul some set of operations

on behalf of a user or another program, with some degree of independence or

autonomy, and in so doing, employ some knowledge or representation of the

user’s goals or desires.” (the IBM Agent),

* “An autonomous agent is a system situated within and a part of an

environment thal senses thal environment and acts on il, in pursuit of ils own

agenda and so as to effect what it senses in the future.” (Franklin, Gasser,

1997)

16