Multi-agent Modelling for a Regulation Support System

of Public Transport

Nabil Morri

1,2

, Sameh Hadouaj

1,2

and Lamjed Ben Said

Emirates College of Technology, Millennium Tower, Sheikh Hamdan Street,

P. O. Box: 41009, Abu Dhabi, U.A.E.

Laboratory SOIE, Tunis University, ISG Tunis, 41 Liberty Street, City Bouchoucha 2000 Bardo, Tunis, Tunisia

Keywords: Multi Agent System, Decision Making, Regulation Support System, Public Transport.

Abstract: The increasing cost of private transport and the rising pollution of the environment pose serious problems in

society, economy and environment. The public transport has become a major challenge of collective and

daily life. However, to encourage people to use a public transport system, the offered service have to be

with good quality. This paper gives effective solutions to improve the quality of public transport service

provided to users. In this paper, we present a Regulation Support System of Public Transport (RSSPT),

based on Multi-agents approach that allows supervising and regulating a multimodal public transport. Its

purpose is to adjust the vehicle schedules where several disturbances come simultaneously. The adjustment

is based on actual traffic conditions. It covers the major criteria that have to be optimized in a traffic

regulation: punctuality, regularity and correspondence.

1 INTRODUCTION

The regulation of public transport is a complex task,

in which decisions are taken according to the current

state of the road network. The network operators

encounter many difficulties to maintain a consistent

traffic and forward planning: theoretical time table,

and respect the use rules (safety rules, business

rules, commercial rules, etc.). The disturbances must

be identified and corrected following the punctuality

criteria, regularity and correspondence. These

different criteria define the service quality of the

passenger in terms of waiting time at station and the

trip time. To insure these criteria the system must

respect the theoretical times of vehicle at station for

punctuality, respect the time interval between

vehicles of the same line for regularity and respect

the waiting time passenger at transfer station for

correspondence.

Hence, the regulation is the mechanism that leads

to effective action decision after collecting the

necessary information and optimizing the service

quality of passenger in terms of punctuality,

regularity and correspondence. It must also deal with

several kinds of disturbances at the same time

(vehicle breakdowns, absence of drivers, blocked

route, etc.) and cover the multi-modality network.

Our objective is to build Regulation Support

System of Public Transport for multimodal traffic

that is able to supervise and regulate the traffic. Our

model takes into account the major criteria that have

to be optimized in a traffic regulation: punctuality,

regularity, and correspondence (Karim, B., Bonte,

T., Sevaux, M., Tahon, C., 2005). It also deals with

many disturbances at the same times by distributed

decision. The plan of this paper is as follows: section

2 introduces the related works and there limits. In

the section 3 we justify the choice of multi-agent

approach in the modelling of our regulation support

system. Section 4 describes the functional model and

the multi-agents architecture. The section 5 presents

an example of simulation and result. The last section

gives a conclusion and future works.

2 RELATED WORKS

The approaches can be classified into two families:

2.1 Mathematical Approaches

Salah Zidi (Salah, Z., 2007) in his thesis

624

Morri N., Hadouaj S. and Ben Said L..

Multi-agent Modelling for a Regulation Support System of Public Transport.

DOI: 10.5220/0005455906240629

In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 624-629

ISBN: 978-989-758-096-3

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

proposes a technique based on a Support Vector

Machine (SVM) and ant colony algorithms.

However the regulation action acts only on vehicles

passages schedules without considering other

objectives such as correspondence and regularity.

The offered solution is a reconfiguration of new

schedules or routes according to the new traffic

conditions.

Sofiene Kachroudi (Sofiene, K., 2010) proposes

a regulating approach for both private and public

modes on wide urban network. This approach uses

an optimization method for particle swarms. It’s a

simple meta-heuristic implementation. But it doesn’t

address the problems of correspondence and

punctuality.

R. Hartani (Hartani, 95) establishes linear

mathematical models characterizing the vehicle

movement between two successive stations of a

public transport line in a high density. This method

effectively treats punctuality. However, there is no

direct link between the calculated values and their

impact on the modification of vehicles kinematic

values (e.g. position, velocity, acceleration and jerk:

third derived from of position). In addition, the

update of the vehicles time table is not done in real

time and the correspondence is not taken into

consideration.

Mohamed Mahmoud Ould Sidi (Mohamed, M.,

2006) proposes in his thesis a resolution method that

takes adequate measures regulations for each

incident. The method used is based on evolutionary

algorithms with the theory of sub-assemblies and

fuzzy integrals. Nevertheless, his method does not

address punctuality, regularity, and feasibility.

2.2 Approaches based on SMA

2.2.1 Regulation of Traffic Lights

(Sofiane, H., Neïla, B., 2010) (Neila, B., Lotito, P.,

2005) (Neïla B., Flavien, B., Suzanne, P., Mohamed,

T., 2011) The objective of these approaches is to act

on the traffic lights durations to regulate traffic of

private cars and public transport mainly buses. They

only address the traffic lights regulation in a normal

state in order to adjust the regularity criteria. But, the

correspondence and the punctuality are not treated.

Also, they don’t deem a real cause of the disturbance

and don’t address the multi-modality network.

2.2.2 Regulation using Evolutionary

Approaches

Flavien Balbo (Flavien, B., Scema, G., 2000)

propose a multi-agent representation based on

"Property-based Coordination Principle" (PbC). The

objective of this approach is to solve three recurring

problems in the design of solutions related to

knowledge, space-time dimension and the real

environment dynamics. The tests show the

importance of multi-agent representation. However

the three main criteria (punctuality, regularity and

correspondence) are not explicitly covered in this

approach.

Fayech (Fayech, 2003), presents the regulation as

a reallocation of schedules and itineraries for

vehicles affected by the disturbance. This approach

requires Hamiltonian paths to ensure the feasibility

of the allocated itineraries. However, this technique

doesn’t deal with traffic regularity. Furthermore, the

decision to change or allocate new itineraries can

cause problems for the correspondence.

Bouamrane (Karim, B., Bonte, T., Sevaux, M.,

Tahon, C., 2005) presents a regulation model that

details the cognitive activities in the regulation

process. The decision is integrated in an interactive

environment, but it is based only on punctuality.

Laichour (Laichour, 2002), proposes to regulate

only the correspondence problem by using a limited

number of actions.

Soulhi (Soulhi, 2000) proposes a fuzzy model

technique. His model is based on the regulator

experience. This technique provides only synthetic

results and deals only with the punctuality issue.

2.3 Discussion

Most of the existing works have limits for the public

transport regulation:

 They don’t take into account perfectly the major

criteria that have to be optimized in a public

transport regulation: punctuality, regularity and

correspondence.

 The majority of works don’t address the public

multimodal transport (bus, metro and tram)

 The majority of works take account only the

information related to passengers like waiting

time in station, frequency of coming passenger,

destination of passenger, etc. It is difficult to

collect and manipulate this information.

 They can’t handle multiple disturbances

simultaneously.

 They don’t detect on time the disturbance.

 They don’t ensure a follow up of the regulation

action impact in order to update the information

system on real-time and develop its expertise in

regulation.

Hence, our goal is to implement a Regulation

Support System of Public Transport (RSSPT) that

Multi-agentModellingforaRegulationSupportSystemofPublicTransport

625

overcomes theses limits. For this, we rely on a multi-

agent approach. This orientation is explained in the

next section.

3 MULTI-AGENTS APROACH

AND REGULATION SUPPORT

SYSTEM

In this section we describe the regulation public

transport domain characteristics and we show how

each of these characteristics can be handled by a

multi-agent approach. The regulation public

transport is:

 Distributed: The geographical distribution of

information over the network for vehicles and

stations requires distributed agents. Each agent

has its own information in order to reflect the

real current state of the public transport network.

For example, when vehicles breakdown or there

are delays or accidents.

 Dynamic: it is the daily change of information

concerning the management of public transport.

The system must then use agents able to

continually modify their states and automatically

adapt to dynamic changes in the environment.

For example, a vehicle agent can move forward,

slow down, accelerate and negotiate its passage

with other agents like stations.

 Open: The state of public transport networks

changes continuously. E.g. the appearance of

incidents or the number of vehicles becomes

large. Hence the system must manage agents

who can enter and exit freely. For example, add a

new regulator agent when a disturbance appears

and add a large number of vehicle agents to

simulate congestion situation.

 Heterogeneous: The actors of Regulator

Support System are varied. The use of a multi-

agent approach allows modeling agents with

different behaviors and different granularities.

For example, vehicle, station, regulator, etc.

 Complex: this domain requires entities with

complex nature. These entities can reason and

communicate via messages to solve problem. For

example, each regulator can manage and makes a

decision, and each vehicle can communicate with

station to give some information like passage

time.

Therefore, multi-agents system sweets well the

public transport network domain. Hence our

modelling is based on agents able to communicate,

cooperate and negotiate to detect and resolve

disturbances.

4 SUPPORT REGULATION

SYSTEM MODELLING

4.1 Objective

Our goal is to provide a regulation support system of

public transport (RSSPT). This system is based on

Multi-agents approach. It is designed to detect and

regulate disturbance of public transport by taking

into consideration the major criteria for traffic

regulation: punctuality, regularity, and

correspondence. Furthermore, this modelling should

simultaneously treat many disturbances and consider

the multimodal aspect of the system (bus, metro, and

tram) as well as the type of vehicles (passenger,

school, commercial, etc.). The mode of transport is

important to treat the feasibility; also the type of

vehicle is useful to define the significance of the

criteria (punctuality, regularity or correspondence)

in the evaluation disturbance.

Our system is based on the system presented by

Karim Bouamrane, (Karim, B., Bonte, T., Sevaux,

M., Tahon, C., 2005). It is subdivided into three

modules: the disturbance acquisition, the regulation

and the evaluation module.

4.2 The Functional Model

The following figure describes the different phases

in the decision-making process of the regulation (see

figure 1).

In the first phase, the system supervises the

network. It collects information from the operating

support system of the public transport network to

locate vehicles. After that, it detects disturbances

and evaluates its impact on network activity by

computing utility function based on punctuality,

regularity and correspondence. This function is

inspired by the work of Catholijn who presents a

prototype system for negotiation about cars

(Catholijn, Jonker, Treur, 2001).

In the regulation phase, the system analyzes

detected disturbances by defining the type of risk

(risk of a vehicle train and / or the risk of a gap). It

produces diagnostic and the current state of the

network. According to the diagnostic the regulator

establishes solutions following an optimization of

the utility function and takes the adequate decision

for example to increase or to decrease the number of

vehicle provided for part of the network, to

accelerate or decelerate vehicle, etc.

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

626

Figure 1: The functional model of the RSSPT.

Finally, there is the evaluation. The model

chooses the regulation action according to its

feasibility and controls it in public transport

network.

4.3 Multi-agent Architecture

The different agents of our architecture are described

in figure 2.

Figure 2: Multi-agent Architecture of the RSSPT.

Our architecture is based on multi-agent

approach where agents include domain knowledge.

Their interactions are based on the environment

state. To achieve our RSSPT we design agents that

store data from a theoretical time table and reflect

the current state of the network (station and vehicle)

as well as agents that calculate and evaluate

disturbance and solution for the three criteria:

punctuality, regularity and correspondence.

4.3.1 Agent Modelling

 Station Agent: is related to one or more lines.

Each agent must memorize all theoretical times

and real times of vehicles passage. It calculates

continuously the waiting time of each coming

vehicle V

R_v

. It is equal to:

v =

C -

(1)

With T

is the Current time and T

T_v

is the

theoretical time of coming vehicle. After that, it

provides the necessary information to agent

punctuality, regularity and correspondence so

they calculate their criteria value.

 Vehicle Agent: is characterized by mode (bus,

metro or tram), position, speed, capacity, number

of passengers, line which affected, mission

(academic, commercial or passenger), driver,

remaining work time, etc. When a vehicle passes,

it provides to the station a passage time, number

of passengers descended and number of

passengers mounted. Also, it gives continuously

to the operating support system via GPS its

position. This information is stored to enrich the

knowledge base of the system. Also, each agent

vehicle is related to each criteria agent who use

these information to calculate its criteria value.

 Punctuality Agent: calculates its criteria value

as follows:

= (T



)- T

(2)

With



is the remaining time needed to the real

time of vehicle passage. This remaining time

must take into account the traffic state and the

mode of transportation.

 Regularity Agent: calculates its criteria value V

as follows:

= V

behind

(3)

The T

R_v_behind

is the real time of behind vehicle

passage (see figure 3).

When V

value is less than vehicle frequency, it

means that there is a risk of a vehicle train.

Otherwise we deduce that there is a risk of a gap.

Figure 3: Calculation of the regularity criteria value.

Multi-agentModellingforaRegulationSupportSystemofPublicTransport

627

 Correspondence Agent: calculates its criteria

value V

. It is equal to:

= ∑ f

(



i_v_c

)

(4)

Here f

represents the waiting time factor of ith

corresponding vehicle. This factor indicates the

importance weight of the corresponding vehicles

in the calculation of the regulation solution. It is

necessary that ∑ f

= 1. And



i_v_c

represents

waiting time of ith corresponding vehicle. It is

equal to:



i_v_c

= V

– T

i_v_c

(5)

With T

i_v_c

represents the real time of the ith

corresponding vehicle passage to the transfer

station.

 Regulator Agent: Each vehicle has regulator

agent that continuously receives the criteria

values calculated by the agent’s punctuality,

regularity and correspondence. Then the

regulator calculates the utility function according

to three criteria (punctuality, regularity and

correspondence).

U= ∑ (W

) with O={p,r,v} (6)

Here the weights W

are relative to the

importance of the different criteria. E.g.

punctuality for school bus is more important than

the passenger bus, against keeping good

correspondence is more interesting for passenger

bus than school bus. The weights values are

between 0 and 1 and It is necessary that

∑ W

=1 with O={p,r,v}.

4.3.2 Dynamic Modelling

We describe dynamic view of our model by using

the sequence diagram (see figure 4).

Each station knows the position station from the

operating support system and calculates the waiting

time from each coming vehicle and sent it to criteria

agent. So, each criteria agent calculates its criteria

value and sent it to the regulator agent. After that,

the regulator agent calculates the utility function. If

the value of this function falls into a critical values

space, the regulator starts the regulation phase. In

this step the regulator use its expertise to optimize

the utility function and produce a hierarchical list of

feasible decision. This list is inspired from the work

of Z. Salah (Salah, Z., 2007). After that it sent this

decision list to the operator. The operator maximizes

the solution satisfaction degree and respects the

decision feasibility. The value of this function must

be as minimal as possible. It is the value of the

theoretical utility function.

Figure 4: Sequence diagram of the RSSPT.

At the end, the operator chooses the appropriate

regulation action decision from this list. After that,

the system updates its knowledge base, and updates

the information of operating support system. We

note that each vehicle has a regulator that operates

independently of other regulators.

A unique characteristic of this model is that the

same agents and data are used to ensure the three

phases of our RSSPT: acquisition, regulation and

evaluation. This makes the model simpler and solves

the problem of data duplication.

In addition this architecture detects the

disturbance on time and ensures self-organization

among the different agents to achieve a collective

goal. Furthermore, the regulation can deal with

many disturbances at the same time and the solution

is taken into account a real time.

5 SIMULATION AND RESULT

We applied this work to scenario based on a real

transportation system existing in Abu Dhabi. We

used an algorithm with a line 008 which has

frequency bus 20 minutes (regularity). A disruption,

due to a traffic accident between two cars, slows

down bus. The delay of the disrupted bus at its

arrival to the Bateen station is estimated 7 minutes:

We estimate that there are 4 minutes for the coming

bus to arrive at station (



Rt=

4) and 3 minutes that the

coming bus not comes yet at 10h: 33min. The

disrupted bus has two correspondences lines 009 and

010. We assume that there is no disturbance on the

behind bus. To calculate the correspondence criteria,

we assume that all corresponding bus have the same

importance in the transfer station (f

=0.5 and f

=0.5).



i=1

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

628

According to the theoretical time table presented

in the following table (see table 1), each criteria

agent calculates its criteria value as following: V

=7,

=27 and V

=15. The regulator agent calculates also

the utility function: U=16.86. We want to find

regulation that minimizes this value. Note that the

optimal value is equal to theoretical value=8.58.

Table 1: Times of bus passage at Bateen station.

Line Theoretical

time at station

Real time

at station

Frequency

008 10h: 30min 10h: 37min 20

009 10h: 23min 10h: 23min 20

010 10h: 25min 10h: 25min 20

After analysis of the situation, the system

recommends that the better action regulation is a

deviation: change the road of the disruption bus

without skip stations. In this case the disrupted bus

comes 3 minutes earlier and the utility function

value becomes equal to 10.23 with V

=4, V

=24 and

=3. We deduce that the utility function is simple to

use and a good way to optimize the regulation

action.

6 CONCLUSION AND FUTURE

WORKS

In this paper, we presented a Regulation Support

System of Public Transport (RSSPT) based on a

multi-agent approach. Unlike other works, our

model takes into account the punctuality, regularity

and correspondence. These criteria are detected and

regulated at the same time when several disturbances

appear simultaneously. Moreover, this RSSPT deals

with multimodal traffic of public transport

The objective of this research is to improve the

passenger quality service of public transport. In our

approach, regulation of public transport is

distributed and obtained thanks to communication,

collaboration and negotiation between

heterogeneous agents. Firstly, we explained shown

that existing approaches of support system

regulation present several limits. Secondly, we have

presented how we used multi-agents approach in our

modelling. Thirdly, we described our multi-agent

strategy that computes utility function on witch

making decision is based. This function is based on

the real condition of traffic situation and takes into

account criteria: punctuality, regularity and

correspondence.

The first experimentation shows that our RSSPT

can sort the different feasible regulation actions

according to the utility function value. The

implementation is in progress using a multi-agents

platform JADE. A deeper validation will be done

using with a real value of the traffic network.

REFERENCES

Flavien, B., Scema, G., 2000. Modélisation d’une

perturbation sur un réseau de transport : le modèle

Incident, IAE – 15.

Flavien, B., Suzanne, P., 2005. Decision Support Systems,

Volume 41, Issue 1.

Neïla, B., Flavien, B., Suzanne, P., Mohamed, T., 2011.

IEEE/WIC/ACM, International Conferences on Web

Intelligence and Intelligent Agent Technology, wi-iat,

vol. 2, pp.

Flavien, B., 2010. Modélisation d’une perturbation sur un

réseau de transport : le modèle Incident, HH Cheng -

Intelligent Transportation Systems, IEEE.

Salah, Z., 2007. Thesis, SARR: Système d’Aide à la

Régulation et la Reconfiguration des Réseaux de

Transport Multimodal.

Mohamed, M., 2006. Thesis, Contribution à l’amélioration

des systèmes d’aide à la décision pour la régulation du

trafic dans les réseaux de transport collectif.

Neila, B., Lotito, P., 2005. An intermodal traffic control

strategy for private vehicule and public transport. 10th

Euro Working Group on Transportation Poznan

Poland.

Sofiane, H., Neïla, B., 2010. Régulation du trafic urbain

avec une priorité aux transports en commun a l’aide

d’un système multi-agents, 8e Conférence

Internationale de MOdélisation et SIMulation -

MOSIM’10.

Sofiene K., 2010. Thesis, Commande et optimisation pour

la régulation du traficurbain multimodale sur de

grands réseaux urbains.

Fayech, 2003. Thesis, Régulation des réseaux de transport

multimodal : Systèmes multi-agents et algorithmes

évolutionnistes.

Laichour, 2002. Thesis, Modélisation Multi-agent et aide

à la décision: application à la régulation des

correspondances dans les réseaux de transport urbain.

Hartani, 95. Thesis, Modélisation des systèmes fous :

Contributions théoriques et applications.

Catholijn, Jonker, Treur, 2001. An Agent Architecture for

Multi-Attribute Negotiation. In Nebel, B., 17th

International Joint Conference on AI, IJCAI’01.

Karim, B., Bonte, T., Sevaux, M., Tahon, C., 2005. SART:

un système d’aide à la décision pour la régulation d’un

réseau de transport bimodal, LAMIH/SP, UMR-CNRS

8530.

Multi-agentModellingforaRegulationSupportSystemofPublicTransport

629