Multi-Agent Approach for Evacuation Support System

Shohei Taga

, Tomohumi Matsuzawa

, Munehiro Takimoto

and Yasushi Kambayashi

Department of Computer and Information Engineering, Nippon Institute of Technology,

4-1 Gakuendai, Miyashiro-machi, Minamisaitama-gun, Saitama 345-8510, Japan

Department of Information Sciences, Tokyo University of Science, 2641 Yamazaki, Noda 278-8510, Japan

Keywords: Multi Agent, Mobile Agent, Ant Colony Optimization, Mobile Ad Hoc Network, Contingency Plan, Risk

Management.

Abstract: We propose a system that supports evacuation after a large-scale disaster. When a large-scale disaster

occurs, collecting information by using portable devices is difficult, because communication base stations

collapse and traffic congestion occurs. Evacuees are at a loss where they should go for safe places in lack of

information. In order to overcome these problems, we have proposed and evaluated a multi-agent system

that is built on MANET. Our aim is to let the users share information on MANET, and provide safe routes

to the destination by using collected information. In the previous paper, we proposed and implemented the

main functions of the proposed system, and performed feasibility study of the multi-agent system by using

simple simulator. In this paper, we report the revised system and investigate the relationships between the

number of the users and diffusivity of information, and the number of the generated mobile agents by using

newly constructed simulator. In the experiments for evaluation, we simulated a realistic situation by using

real map data and we took stochastic change of the situation into consideration, because the situation must

be getting worse by time elapses.

1 INTRODUCTION

When a large-scale disaster, such as big earthquakes

occurs, a large-scale fire occurs, buildings collapse,

and roads break. In addition, the public transports

may paralyze and large-scale traffic congestion may

occur. As a result, a number of evacuees may

inundate the damaged towns. Furthermore, people

starts to use the communication devices, such as

smartphones, to collect information about the service

status of public transport and safety information of

family and friends. As a result, communication base

stations may become too congested as well as

destroyed. The users of communication services

cannot depend on the traditional communication

infrastructure, because of the collapse of the base

stations and extreme congestion.

When the Great East Japan Earthquake occurred

in 2011, we experienced shaking of magnitude 5

even in Tokyo. Tokyo

is more than 300km south from

the seismic centre. About 5.15 million people could

not confirm the safety of their family members due

to communication network congestion. A cascading

communication failure prevented people from

contacting their families and friends. People had to

go back home using unfamiliar roads on foot. In

such situations, where the buildings may be on fire

and bridges may collapse, it is very dangerous to

walk on the roads without navigating information.

Some companies have already developed support

applications that address large-scale disasters. An

example is the "Disaster return home support map"

that Mapple-on, Inc. offers. (Mapple ON Co., Ltd.,

2016). This application installs the map data to the

users’ portable device in advance. Even if a

large-scale disaster cut off the Internet, people can

use the map information off-line. This application

provides the minimum necessary information to cope

with large-scale disasters such as alternating roads

and emergency shelter. This application also

provides the route guidance function to the

destination of each user. However, the system lacks

of a function to collect the environmental

information that dynamically changes, such as fires

or buildings collapsing after the disaster. Therefore,

a situation may happen such that the system suggests

a safe route that is actually damaged and impassable,

and of course dangerous. Thus, we need a means to

provide alternative routes to the destination for each

user by collecting environmental information about

the dynamically changing post disaster situation.

220

Taga S., Matsuzawa T., Takimoto M. and Kambayashi Y.

Multi-Agent Approach for Evacuation Support System.

DOI: 10.5220/0006194202200227

In Proceedings of the 9th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2017), pages 220-227

ISBN: 978-989-758-219-6

We proposed a system that supports pedestrians

returning home after a large-scale disaster (Taga et

al., 2016). The system was intended to provide an

appropriate route to the destination for each user. We

designed the system that can collects information

even when a communication failure occurs, and

generate routes dynamically based on the collected

information. In the previous paper, we proposed and

evaluated the main functions of the system, and

performed a feasibility study of the multi-agent

system by using simple simulator. In the

experiments, we observed that the communication

load to the agent system is unrealistically high. For

that reason, we have simplified the agent system. In

this paper, we report the revised system and

investigate the relationships between the number of

the users and diffusivity of information, and the

number of the generated mobile agents by using

newly constructed simulator. In the experiments for

evaluation, we have simulated realistic situation by

using real map data and taking stochastic change of

the situation into the consideration, because the

situation must be getting worse by time elapses.

It is preferable to build a mobile ad hoc network

(MANET) by direct communication between

portable devices to cope with the post disaster

circumstance. Since the network uses portable

devices as relays, communication can continue

without the intervention of the base station. Since

MANET is a network constructed by using only

portable devices, it is more prone to disconnection of

the communication networks than the station-base

communication system. Furthermore, the proposed

system is supposed to be used by evacuees after

large scale disasters, therefore the users of the

system as communication nodes are always moving,

this problem requires even more serious

consideration on routing.

In order to mitigate the above problem, we have

adopted mobile agents in our proposed system.

Mobile agents are software agents that hold not only

data but also computing capability when moving.

Thus, mobile agents can continue processing state

before moving on the destination so that they can

decide autonomously and intelligently to where it

should move without the assistance of routers. The

communication is established only at the moment the

mobile agents to move. Therefore, it should be

useful in the communication based on an unstable

MANET environment.

The structure of the balance of this paper is as

follows. The second section describes the

background and discusses the related works. The

third section describes the multi-agent system that is

the basis of our proposed system. In order to

demonstrate the feasibility of our system, we have

constructed a simulator for this system and have

conducted numerical experiments. We report the

results and discuss the future works in the fourth and

fifth sections. Finally, we conclude our discussion in

the sixth section.

2 BACKGROUND

Asakura et al. investigated the calculation of the

evacuation routes after a large-scale disaster. They

have proposed a method that uses Ant Colony

Optimization (ACO), and have shown it is useful in

a simulator. (Asakura et al., 2013).

Alejandro Avilés et al. investigated how to

support people escaping to the emergency exit from

the building (Aviles et al., 2014). In their study, they

assumed some of evacuees use portable device based

guidance system like in our study. They have also

adopted MANET and the ACO in order to derive

escape routes.

ACO is an optimization technique that mimics

the foraging behavior of ants. Ants go back and forth

between the feeding grounds and nest in order to

bring food from feeding areas to nests. At that time,

ants put down volatile chemical substance called

pheromone to the routes they went. Other ants that

are back and forth feeding grounds and nest follow

the pheromone and replenish drop the pheromone.

By these actions, the long paths to the feeding

grounds lose their pheromone by evaporation before

the pheromone replenished. On the other hand, other

ants strengthen shorter path pheromone before

evaporation. As the result, ants derive the optimum

route to feeding area. We call methods that are using

this characteristic ACO. (Dorigo et al., 1996).

In Avilés’s study, an evacuee who succeeded to

escape generates a mobile agent called Goal Agent

(GA) from his or her portable device. Generated GA

repeats the move to the portable devices of other

evacuees, to record the number of times of the

movement of its own to the destination portable

device. Further, an evacuee who is seeking safe exits

generates the other kind of mobile agents called

Routing Agent (RA) from his or her portable devices.

Generated RA repeatedly migrates to the portable

devices in the other evacuees, collects the number of

times of the movement of GA that is recorded in the

destination device. The system uses the number of

movements of the GA as pheromone of ACO to

guide evacuees to the safe route. They proposed a

method of providing the evacuees the trajectory of

Multi-Agent Approach for Evacuation Support System

221

the movement of the smallest number of movements

of GA that should be the optimal escape route.

There is a network technique called Delay

Tolerant Network (DTN). DTN is a technique that

was designed for adapting difficult circumstances to

maintain the network communication, such as nodes

constructing the network are scarce and always

moving. When a source node cannot communicate

with the destination node to which it wants to send

data, the node keeps storing the data to transmit.

When the target node is approaching, it then

transmits the previously accumulated data. When

sending data to a distant node, the system treats the

nodes between them as relays and achieves the data

transmission like a bucket brigade. Thus, if

temporarily communication is established

intermittently, it is possible to communicate between

arbitrary nodes. This feature is similar to that of

mobile agents.

Nishiyama et al. investigated communication

network that combines the DTN and the MANET by

using only smartphones (Nishiyama et al., 2016).

They constructed a communication network with

multi-hop connections between the smart phones by

Wi-Fi. As a feature, it utilizes the switching network

techniques with different characteristics of MANET

and DTN in accordance to the communication

conditions autonomously. When the system finds

many nodes, it uses MANET, and it found

communication is difficult, it switches to use DTN.

Thus, the reduction of the communication load and

power saving is achieved. Moreover, by utilizing the

drones in addition to smartphones, additional

scalability can be expected. Our study is also using a

network technique combined with MANET and

DTN by the mobile agents.

Nishiyama et al. focused only on constructing

communication network, while our research focuses

the dynamic routing using mobile agents. For both

MANET and DTN, the communication network

among the portable devices that people have is

constantly changing. In addition, all the evacuees

who use the system are not necessarily looking to the

same destination. They should be able to set the

destination for their own. Mobile agent perceives its

environment, and it can autonomously change the

behavior. Taking advantage of these features, we are

implementing a system that is capable of responding

flexibly to a wide variety, such as user requests and

network environment changes.

The proposed system has two main features; one

is to build a network consisting of only portable

devices by wireless communication. The other one is

the sharing and spread information by the

multi-agent system in the network. We assume

MANET as the communication network. Imaginary

situation in which the proposed system is used

would be a scene where the congestion of the

portable device communication occurs. Therefore,

we can assume that there are many users, and

constructing MANET would be no problem.

However, in our proposed system, a mobile agent

has certain destination and the direction is limited. If

a mobile agent cannot find a destination node in its

limited direction, the mobile agent has to wait. Such

a case needs DTN. Therefore, the network that we

are proposing has both the features of MANET and

DTN. We describe the multi-agent system in detail in

the next section.

3 AGENT SYSTEM

The proposed system is a multi-agent system. A

multi-agent system is a system implemented by

multiple agents that operate cooperatively. In the

agent system, there are two types of agents. One is

static agents that reside on the users’ portable

devices: they are the information agent (IA) and the

node management agent (NA). The other is mobile

agents that move among the users’ smartphones:

they are the information diffusion agent (IDA), the

safety information agent (SA) and the container

agent (CA). The user generates mobile agents as

needed, and mobile agents convey messages while

moving between portable devices. Static agents are

resident in portable devices and process the data

gathered by mobile agents as well as provide user

interface. They also have responsibility to generate

mobile agents. Figure 1 shows the relations between

these agents.

Figure 1: Multi-agents system.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

222

3.1 Information Agent (IA)

IA is a static agent. Its roles are providing the user

interface, generating mobile agents, and deriving the

evacuation routes on the map for each user.

First, we describe the user interface. IA displays

the derived route on the map with useful information

such as impassable points and first aid stations.

When the user finds an impassable point or a first aid

station, he or she presses the corresponding diffusion

button on the screen so that IA creates an IDA to

diffuse the newly discovered information. IA

updates the information on the map as IDAs, which

are created by other users, arrive and give IA new

information.

Second, we describe the generation of mobile

agents. When the generation condition of each

mobile agent is satisfied, IA generates the mobile

agent. When IA generates the mobile agent, it passes

necessary information to the mobile agent, and

dispatches it to one of the portable devices nearby.

When the mobile agent generated by another

portable device arrives, IA receives the information

and passes it to NA.

Third, we describe the route derivation. IA

derives the route to the destination by using the

collected information. While the default destination

is set to the home, the system switches the

destination to the rest area after certain time has

elapsed since system startup, in order to give the user

a break. The user can choose whether the switching

occurs and duration time for switching. IA derives

the return route by Dijkstra algorithm, and then

displays it in the map on the interface screen

(Dijkstra, 1959). If IA receives information of an

impassable point on the current route, it re-derives a

new route that avoids that point. IA also re-derives

the route when the user changes the destination so

that IA provides the latest and secure route.

3.2 Node Management Agent (NA)

NA is a static agent that keeps the collected

information and passes the information to IA on

request. The stored information includes GPS

information representing the current location, the

current destination, the impassable locations, the first

aid stations and the location of the rest areas along

with the time they are stored and the IDs of the IDAs

that convey the information. When redundant

information is delivered, NA deletes the old

information by referring to the time information. In

addition, after a certain time from the store, the old

information is deleted.

3.3 Information Diffusion Agent (IDA)

IDA is a mobile agent that diffuses useful

information such as impassable points on roads and

bridges, resting places and the first aid stations that

the user finds. IDAs collectively implement the

information sharing among the users.

When the user presses one of the diffusion

buttons on the interface screen, IA produces an IDA.

The IDA has the information about what the user

finds the location and the time. Then, the IDA clones

itself to the number of portable devices that are

linked to the originating smartphone. One copy

moves each of the neighbor portable devices. IDA

has a unique ID and all the IDs of the duplicated

IDAs are the same. Upon arrival, IDA passes its

information and its ID to the IA of the portable

devices it arrives to and it clones itself again to all

connected portable devices. If its ID is already stored,

the IDA self-destructs without doing anything. By

doing so, IDA prevents dispersing redundant

information. In addition, each IDA disappears after a

certain number of movements.

IDAs collectively implement the pheromone in

ACO. IDA determines the value of its information

according to the distance between the point of origin

and the current position, the number of movements it

performed, the elapsed time from it was generated.

In other words, IDA reduces the value of its

information as the number of movements, the

elapsed time, the distance between the point of

origin and current position increases. When the

information value gets to be below a certain

threshold, the IDA gets to be useless and

self-destructs (Figure 2).

Figure 2: Information diffusion agent.

Multi-Agent Approach for Evacuation Support System

223

3.4 Safety Information Agent (SA)

SA is a mobile agent that transmits the safety

information. Safety information is not related to the

evacuation. However, safety information of family

members and friends is what people want to know.

SA conveys information such as user’s name,

date of birth, current position, and state (fine or

injured). IA generates SA, and passes the input

information described above. IA generates SA at

predetermined time intervals. Generated SA moves

to the portable devices that are in the direction of the

location registered in advance (such as a home or

office) until it reaches to the destination. If SA

cannot reach to the destination for a long time, SA

self-destructs. When the user cannot move for

reasons such as injury, he or she needs to ask for

help. Such a time, SA moved to the nearest

emergency station, such as police or rescue stations

(Figure 3).

Figure 3: Safety information agent.

3.5 Container Agent (CA)

CA is another mobile agent, which bundle multiple

mobile agents with the same destination together. In

the preliminary experiments, we found the

communication load for mobile agents was too high.

Especially in a large city, it is large enough to

overload the communication capacity. This may

cause problems in unstable communication networks

such as MANET and DTN. The mobile agent may

not be able to move due to network congestion and

battery consumption.

In order to mitigate this problem, we take

advantages of the idea of the container agent system

proposed by (Kambayashi et al., 2016). When IA

finds multiple mobile agents with the same

destination, it create a CA that bundles the mobile

agents move to the same device in one batch and

conveys at once. This method enables multiple

mobile agents to be transferred at once. Once the

transfer completes, CA will be discarded and the

mobile agents included in it will be unbundled into

independent mobile agents.

4 EXPERIMENTS

We have designed and implemented a simulator in

order to evaluate the proposed system. Figure 4

shows the simulator. This shows the situation of a

post disaster city. This is the map of a part of Tokyo,

where the first author lives. We assume the

communication failure has occurred in the city.

4.1 Experimental Conditions

Elements displayed on the simulator are: 1) area map

of the simulated field, 2) the users (circle icon) who

use the proposed system, 3) the communication links

between portable devices owned by the users, 4) the

danger zone that the user cannot passes (x marked),

5) the destinations of the users (areas of “G” icon ).

When the simulation starts, the users are placed

on the field in random positions. The destination for

each user is also randomly selected from safe places.

In Figure 4, there are four safe places as destinations.

When the simulation starts, the entire user begins to

move in unison to their destinations. Each user aims

the destination in the shortest route, because the

system suggests the shortest path. The route search

to the destination uses Dijkstra method.

When a user comes close to the danger zone

while he or she is moving, the user stops moving and

generates IDA. The IDA has the position

information of the danger zone in contact. The IDA

duplicates itself and moves to the portable devices

that are linked by MANET to the portable device

that has created IDA. Then IDA passes information

to IA on the destination portable device, and diffuses

again. This action repeats as long as IDA is in range

from the discovered danger zone within 120m. After

the user generates IDA, the user can turn back the

way he or she has come. Then, IA takes the

discovered danger zone in consideration and

recalculates the route to the destination.

When IA on a user’s portable device obtains the

information of danger zone by arriving IDA

generated by other user, the IA recalculates the route

to the destination that avoids the danger zone. As

time elapses, the number of danger zones increases

as well as existing danger zones spreading by a

predetermined time interval. When a user is caught

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

224

Figure 4: The simulator.

by danger zones and he or she finds no way to go out,

the user dies. If a user dies or arrives to the

destination, the user stops moving. Simulator

terminated when a certain percentage number of

users has arrived at the destination.

In this simulator,

“one round” expresses the unit length the user moves

and process agents. The number of rounds is

recorded and be treated as elapsed time of the

simulation. We recorded the following data: (a) The

maximum number of mobile agents that reside on

one of the portable device, (b) the number of times

that the users touched to the danger zones, and (c)

elapses time to complete the evacuation. We have

conducted experiments with the number of users 100,

200, 300, 400 and 500. For each case, we have

performed ten times and recorded the average

numbers for three data above. We have performed

experiments three cases for (b) and (c). The case 1 is

the case where each portable device can have

unlimited number of mobile agents, the case 2 is the

case where each portable device can have up to five

mobile agents, and the case 3 is the case no user uses

the system. Map used in the simulation is Arakawa,

Tokyo where the first author lives. The simulation

field is

900m * 1200m, and the

communication range

is set to

40m. That

is the distance that most Japanese

Wi-Fi devices can reach in general. Each simulation

ends when

90 % of users reach to their destination.

4.2 Experimental Results

The experimental results are as follows. The

maximum number of mobile agents was proportional to

number of users

Figure 5 shows the number of times that the

users touched to the danger zones. The graphs

display the three cases of the maximum number of

residing mobile agents: no limit, five, and zero.

Naturally, the more mobile agents can move and

reside on portable devices, the less the number of

times that the users touched to the danger zones.

Figure 5: The number of times that the users touched to the

danger zones.

Multi-Agent Approach for Evacuation Support System

225

Figure 6 shows the duration time to complete the

evacuation for each of three cases.

Figure 6: Elapses time evacuation to complete.

4.3 Observations

First, let us consider the maximum number of mobile

agents that resides on each portable device. The

number of agents is a little more than proportional to

the number of the users. When the number of users

is 100, the number of mobile agents on one portable

device is at most ten. When the number of users is

500, however, the number of agents on one portable

device becomes more than 70. Those mobile agents

are coming in and going out on each portable device.

Kambayashi et al. investigated how long it took one

mobile agent to move between portable devices by

using IEE802.11. (Kambayashi et al., 2016).

According to their study, it takes about 1.6 seconds

for one mobile agent to move between two portable

devices. Even though their study did not use as many

mobile agents as our study, the result suggests it

would take more than several tens of seconds for

movements of 70 mobile agents. In such a case, such

process would cause battery consumption and

congestion by mobile agents.

Next, let us consider the number of times that

users touched the danger zones. We observed that

using our system drastically decreases the number of

touches to the danger zones as shown in Figure 5. In

the case of 100 users, our system decreases the

number of touches to danger zones by almost 50%.

In the case of 500 users, the effect is further dramatic.

In the case of no limit on the maximum number of

mobile agents, the number of touches to danger

zones decreased to about 1/7 as compared with those

of not using the proposed system. This result

suggests that, the more the participants in the

network constructed by the proposed system, the less

people face to the dangers due to improved

information propagation.

As described above, however, if there were 500

users, there would be too many mobile agents. Too

many mobile agents effectively paralyze the

multi-agent system due to congestion. Therefore, we

consider the case of the limited number of mobile

agents and the case of not-limited number of mobile

agents. The maximum number of mobile agents in

the case of not-limited number is about seventy. The

maximum number of mobile agents in the case of

limited number of agents is five. Even though we

limit the number of mobile agents that can reside on

one portable device to five, we can decrease the

number of times the users touch to the danger zones

drastically. When people do not use the system,

almost all the participants touch to the danger zones.

On the other hand, people use the system with

restricted portable devices, i.e. the number of mobile

agents is up to five, touch to the danger zone only

one third times.

Finally, let us consider the duration time to

complete the evacuation. Whether the number of

mobile agents is limited or not limited does not

affect very much, while whether using the system or

not affects very much. Figure 6 clearly show the

effectiveness of our system. People using the system

can move to their safe places much quicker than

people not using the system can.

5 DISCUSSIONS

As shown in Figure 5, the more the participants in

the MANET, the better the performance of the

proposed system. However, increasing number of

mobile agents leads to various problems. In order to

cope with these problems, we have limited the

number of mobile agents that can reside on one

portable device in the experiment (b). Our results

suggest that even though limiting the number of

mobile agents decreases the performance of the

proposed system, the effect is not fatal. Limiting the

number of mobile agents can help to construct a

system that is more realistic. In this study, we only

experimented with the maximum number of the

agents five, but it is necessary to propose a

theoretical basis for such restrictions.

As an example, we are reconsidering the way to

dispatch IDAs. This agent performs the role of

diffusing the discovered information to other users.

It is not necessary to move IDA to a user who is

closer to the discovered information than those who

discover the information. Because, this user can find

the discovered information before IDA arrives to this

user. This problem requires further consideration.

Due to the nature of the proposed system, false

information might be diffused. In fact, when the

Great East Japan Earthquake of 2011 occurred, there

were problems with false information diffused

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

226

through social networking service. In time of a

large-scale disaster, information is of high

importance. In addition, many people become

disoriented and can be deceived easily by the false

information. Therefore, it is necessary to take

precautions against this.

One of the ways to address this problem may

take advantages of ACO. The characteristic of the

pheromone in ACO applies to diffused information.

For example, the system can treat old information as

less important than new information. Then it

becomes possible to select and discard information.

It is not clear, however, how to set the pheromone

values. Goto et al. have studied a route search using

ACO. They have used two types of pheromones.

One pheromone calculates the escape route. Another

pheromone deletes the pheromone, which exists in

the danger zone. From these pheromones, the system

calculates routes to avoid the danger zone. (Goto et

al., 2016).

6 SUMMARY

In this paper, we proposed a system that supports

evacuation at the time of large-scale disasters. In

order to cope with communication failure due to

damage and congestion of the communication base

station, we proposed to build a MANET via

communication between portable devices, and to

collect information by a multi-agent system. We

have implemented a simulator that evaluates how

much the proposed system can save evacuees at the

time of large-scale disasters. On the simulator, we

have performed many experiments and recorded

three data: (a) The maximum number of mobile

agents that reside on one of the portable device, (b)

the number of times that the users touched to the

danger zones, and (c) elapses time to complete the

evacuation. We have found that the more people join

the MANET, the better the information spreads,

though having too many mobile agents also leads to

problems.

In addition to the problem with over-proliferation

of the mobile agents, the current system also suffers

from the problem with diffusing false information.

There is certainly needs to improve the simulator for

a more realistic simulation. For example, Goto et al.

created a simulator based on the real tsunami data of

Rikuzentakada after the Great East Japan Earthquake

occurred in 2011 (Goto et al., 2016). Ushiyama et al.

reproduce the details of this tsunami phenomenon

from various recorded data and testimony (Ushiyama

et al., 2012). We are planning to use this data.

ACKNOWLEDGEMENTS

This work is supported in part by Japan Society for

Promotion of Science (JSPS), with the basic research

program (C) (No. 25330089 and 26350456),

Grant-in-Aid for Scientific Research.

REFERENCES

Aviles, A. M., Takimoto, M., Kambayashi, Y., 2014.

Distributed evacuation route planning using mobile

agents. Transaction on Computational. Collective

Intelligence XVII, LNCS 8790, pp.128-144.

Taga, S., Matsuzawa, T., Takimoto, M., and Kambayashi,

Y., 2016. Multi-Agent Approach for Return Route

Support System Simulation. Proceeding of the Eighth

International Conference on Agents and Artificial

Intelligence, vol. 1, pp.269-274.

Mapple ON Co., Ltd., 2016. Disaster return home support

map. http://mapple-on.jp/products/kitaku.

Nishiyama, H., Ito, M., Kato, N., 2014.

Relay-by-Smartphone: Realizing Multi-Hop

Device-to-Device Communications. IEEE

Communications Magazine, vol. 52, no. 4, pp. 56-65,

Apr.

Goto, H., Ohta, A., Matsuzawa, T., Takimoto, M.,

Kambayashi, Y., and Takeda, M., 2016. A Guidance

System for Wide-area Complex Disaster Evacuation

Based on Ant Colony Optimization. Proceeding of the

Eighth International Conference on Agents and

Artificial Intelligence, pp.262-268.

Kambayashi, Y., Nishiyama, T., Matsuzawa, T., and

Takimoto, M., 2016. An Implementation of an Ad hoc

Mobile Multi-Agent System for a Safety Information,

Proceeding of the Thirty-sixth International

Conference on Information Systems Architecture and

Technology, AICS vol. 430, pp.201-213.

Ushiyama, M., and Yokomaku, S., 2012. Estimation of

situation in Rikuzentakata city just before tsunami

attack based on time stamp data. Japan Society for

natural disaster science, Vol. 31, No. 1, pp. 47-58.

Asakura, K., Fukaya, K., and Watanabe, T., 2013.

Construction of navigational maps for evacuees in

disaster areas based on ant colony systems.

International Journal of Knowledge and Web

Intelligence, Vol. 4, pp. 300-313.

Dijkstra, E. W., 1959. A Note on Two Problems in

Connexion with Graphs. Numerische Mathematik,

vol.1, pp. 269-271.

Dorigo, M., Maniezzo, V., Colorni, A., 1996. Ant system:

Optimization by a colony of cooperating agents, IEEE

Transaction on System, Vol. 26, No. 1, pp.29-41.

Multi-Agent Approach for Evacuation Support System

227