C. Ramos et al. (eds.), Ambient Intelligence - Software and Applications,  
Advances in Intelligent Systems and Computing 291,  
241
DOI: 10.1007/978-3-319-07596-9_27, © Springer International Publishing Switzerland 2014 
 
Multi-agent Technology to Perform Odor Classification  
Sigeru Omatu1, Tatsuyuki Wada1, Sara Rodríguez2,  
Pablo Chamoso2, and Juan M. Corchado2 
1 Osaka Institute of Technology, Osaka, Japan 
2 Computer and Automation Department, University of Salamanca, Spain 
omatu@rsh.oit.ac.jp, coco.tk.family@gmail.com,  
{srg,chamoso,corchado}@usal.es 
Abstract. Quartz crystal microbalance (QCM) sensors are used to measure and 
classify odors. In this paper, we use seven QCM sensors and three kinds of 
odors. The system has been developed as a virtual organization of agents using 
the agent platform called PANGEA (Platform for Automatic coNstruction of 
orGanizations of intElligents Agents), which is a platform to develop open mul-
ti-agent systems, specifically those including organizational aspects. The main 
reason that justifies the use of the agents is the scalability of the platform; that 
is, the way in which it models the services. The functionalities of the system are 
modeled as services inside the agents, or as SOA (Service Oriented Approach) 
architecture compliant services using Web Services. In this way, it is possible to 
improve odor classification systems with new algorithms, tools and classifica-
tion techniques. 
Keywords: Odor sensing, odor classification, multi-agent systems, virtual or-
ganizations, QCM sensors. 
1 Introduction 
During the last years, major advances have been made in the field of Ambient Intelli-
gence [1], [2], which has come to acquire significant relevance in the daily lives of 
people [5], [6], [7]. Ambient Intelligence adapts technology to people’s needs by pro-
posing 3 concepts: ubiquitous computing, ubiquitous communication and intelligent 
user interfaces. The development of new frameworks and models to allow informa-
tion access, independently of the location, is needed in order to achieve these targets. 
Wireless sensor networks [3], [4], [22], provide an infrastructure, which is able to 
distribute communications in dynamic environments by incrementing mobility and 
efficiency independently of the location. Sensor networks interconnect a large amount 
of sensors and manage information in the intelligent environment. Many times infor-
mation management is done in a distributed way. However, it is necessary to have 
distributed systems with enough capabilities to manage sensor networks in an effi-
cient way and to include elements with some degree of intelligence that can be  
embedded in the devices and act both autonomously and in coordination with the 
distributed system. Multi-agent systems are a suitable alternative to perform this type 
of systems. 
242 S. Omatu et al. 
There are several proposals to build smart environments that combine multi-agent 
systems and sensor networks [8], [9], [10], [11], [12], [13] , [14], [15], [16], [17], 
[18], [19], [20], [21]. New approaches are needed to support evolutional systems and 
to facilitate their growth and runtime updates. The dynamics of open environments 
have promoted the use of Virtual Organizations of Agents (VOs). A VO [25], [26], 
[27], [28], [29] is an open system designed for grouping; it allows for the collabora-
tion of heterogeneous entities and provides a separation between the form and  
function that define their behavior.  However, it is not possible to find an existing 
multi-agent architecture to work on the concept of virtual organizations and to pro-
vide agents capable of working with any type of sensor or device. This article consid-
ers different types of odor sensors and aims to classify odors according to sensing 
data by using quartz crystal microbalance (QCM) sensors. QCM sensors are sensitive 
to odors and allow the precise measurement of odor data. Using many QCM sensors, 
we will attempt to classify various kinds of odors based on neural networks. To model 
the system, virtual organizations of agents, which are capable of bringing a greater 
number of possibilities, are presented. These agents are connected with PANGEA 
[23], a multi-agent platform designed on the basis of virtual organizations, aimed at 
the creation of intelligent environments. 
Over the last decade, odor-sensing systems (called electronic nose (EN) systems) 
have undergone important developments from a technical and commercial point of 
view. EN refers to the ability to reproduce the human sense of smell by using sensor 
arrays and pattern recognition systems [30].  
The authors in [31] present a type of an EN system to classify various odors under 
the various densities of odors based on a competitive neural network by using learn-
ing vector quantization (LVQ). The odor data were measured by an odor sensor array 
made of MOGSs. We used fourteen MOGSs of FIGARO Technology Ltd in Japan. 
We considered two types of data for classification in the experiment. The first type 
included four types of teas, while the second included five types of coffees with simi-
lar properties. The classification results of teas and coffees were approximately 96% 
and 89% respectively, which was much better than the results in [32], [24]. 
The article is structured as follows. First, the PANGEA platform is described in 
section 2, detailing the structure of the virtual organizations used in the odor classifi-
cation case study. Both the platform and virtual organizations are evaluated in a case 
study consisting of an intelligent environment for odor recognition. Finally the results 
of the case study and the conclusions reached from this research are presented. 
2 Case Study: Development of a VO for Odor Classification 
This central section of the article presents the integration of the system and the sen-
sors used in the multi-agent architecture, and explains the main concepts of QCM 
sensors. In addition, an overview of the odor sensing system and the measures of odor 
data used are described. 
 
 Multi-agent Technology to Perform Odor Classification 243 
2.1 Integration in a Multi-agent Platform (PANGEA) 
With the development of ubiquitous and distributed systems, it is interesting to have 
new agent platforms that facilitate the development of open agent-architectures that 
can be deployed on any device. PANGEA [23] is an agent platform based on organi-
zational concepts. It can model and implement all kinds of open systems, encouraging 
the sharing of resources and facilitating control of all nodes where the different agents 
are deployed. 
It is essential to have control mechanisms that enable new devices to be included in 
a single platform where they can be easily integrated, managed and monitored. In this 
case PANGEA, with its model of agents and organizations, provides the necessary 
features to function as the base platform when developing a comprehensive system. 
In order to facilitate control of the organization, PANGEA has several agents that 
are automatically deployed when starting the platform operation: OrganizationMa-
nager and OrganizationAgent are in charge of the management of the organizations 
and suborganizations; InformationAgent is in charge of accessing the database con-
taining all pertinent system information; ServiceAgent is in charge of recording and 
controlling the operation of services offered by the agents; NormAgent is in charge of 
the norms in the organization; and CommunicationAgent is in charge of controlling 
communication among agents, and recording the interaction between agents and or-
ganizations.  
In addition to the intrinsic PANGEA agents, the organizations developed in the 
present system are the following: 
─ Odor-recognition sensors organization. In this organization all agents belonging to 
an individual odor recognition system is deployed. Such agents may also be of dif-
ferent types (sensor agents, interface agent and identifier agent).  
─ Sensor control central organization. In this organization the agent interface type is 
included, representing each of the odor-recognition sensors organizations together 
with an adapter agent. 
Communication in this case is restricted only to the existing agents in the same or-
ganization, in addition to the control agents that the PANGEA platform offers (as is 
the case of the Information Agent, which accesses the database). 
Each type of agent is engaged in a well-defined task, as explained below: 
─ Sensor agent. It is exclusively dedicated to performing sensor readings and provid-
ing the latest value when an authorized agent requires such data. 
─ Identifier agent. Its function is to perform the necessary calculations for the identi-
fication of odors. It makes use of the ability to communicate with the sensor 
agents, which require the data needed to perform these calculations. 
─ Interface agent. This kind of agent is present in the two types of virtual organiza-
tions cited. It is responsible for providing a communication link with the agents 
outside their own organization of odor-recognition sensors that are authorized to 
establish two-way communications using the appropriate communication format. 
244 S. Omatu et al. 
─ Classifier agent. This a
algorithms described in
Classification Method o
platform. We can say tha
fication of odors by mak
─ Adapter agent. This type
trol. Its function is to tr
each of the associated se
all recognition systems p
source of knowledge. 
Fig. 1. Structure o
2.2 Algorithms: Classif
We will show two types o
based on error back-propag
tion (LVQ).  
First, we will explain th
odors, we adopt a three-lay
method, as shown in Fig.2. 
the gradient method, is give
 
• Step 1. Set the i
• Step 2. Specify 
ponding to the i
• Step 3. Calculat
݊݁ݐ௝ ൌ ෍
ூ
௜ୀଵ
gent performs classification services that implement 
 the following section. To perform the classification
f Odor Data (e-BPNN) is used and implemented on 
t the classifier agent could use new methods for the cla
ing the system scalable in terms of functionality. 
 of agent is in the central organization of the sensor c
y to correct the differences between the measurements
nsors to the sensor agents. Thus, a joint database amo
articipating in the architecture is achieved, expanding 
 
f virtual organizations of the case study in PANGEA 
ication Method of Odor Data (e-BPNN) 
f neural networks: one is a multi-layered neural netw
ation method and the other is a learning vector quanti
e multi-layered neural networks. In order to classify 
ered neural network based on the error back-propagat
 The error back-propagation algorithm, which is based
n by the following steps. 
nitial values of ݓ௝௜, ݓ௞௝, ߠ௝, ߠ௞, and η(> 0). 
the desired values of the output ݀௞, ݇ ൌ 1,2, ڮ , ܭ corre
nput data x୧, i= 1,2, . . . , I in the input layer. 
e the outputs of the neurons in the hidden layer by 
ݓ௝௜ݔ௜ െ ߠ௝ , ௝ܱ ൌ ݂൫݊݁ݐ௝൯, ݂ሺݔሻ ൌ
1
1 ൅ ݁ି௫ 
the 
, a 
the 
ssi-
on-
 of 
ng 
the 
ork 
za-
the 
ion 
 on 
s-
 • Step 4. Calculat
݊݁ݐ௞ ൌ ෍ ݓ
௄
௞ୀଵ
• Step 5. Calculat
• Step 6. Use th
squares of the e
• Step 7. If E is s
weight by the fo
Δݓ௞௝ ؠ ݓ௞௝ሺݐ
Δݓ௝௜ ؠ ݓ௝௜ሺݐ ൅
• Step 8. Go to St
 
Using the above recursi
Fig. 2. Three laye
The neural network (Fig
and output layer k. When 
layer, we can obtain the o
desired value ݀௞  which h
 
 
Multi-agent Technology to Perform Odor Classification 
e the outputs of the neurons in the output layer by 
௞௝௜ ௝ܱ െ ߠ௝ , ܱ௞ ൌ ݂ሺ݊݁ݐ௞ሻ, ݂ሺݔሻ ൌ
1
1 ൅ ݁ି௫ 
e the error en and generalized errors by 
݁௞ ൌ ݀௞ െ ܱ௞ 
ߜ௞ ൌ ߜ௞ܱ௞ሺ1 െ ܱ௞ሻ 
ߜ௝ ൌ ෍ ߜ௞
௄
௞ୀଵ
ݓ௞௝ ௝ܱሺ1 െ ௝ܱሻ 
e following formula to calculate half of the sum of 
rrors in the output of all. 
E ൌ 12 ෍ ݁௞
ଶ
௄
௞ୀଵ
 
ufficiently small, exit the learning. Otherwise, modify 
llowing equation: 
൅ 1ሻ െ ݓ௞௝ሺݐሻ ൌ ߟߜ௝ ௝ܱ௞        ֚  ݓ௞௝ ൅ Δݓ௞௝  
1ሻ െ ݓ௝௜ሺݐሻ      ൌ ߟߜ௜ ௝ܱ            ֚  ݓ௝௜ ൅ Δݓ௝௜  
ep 3. 
ve procedure, we can train the odor data. 
  
red neural network with the error back-propagation 
. 2) consists of three layers: input layer i, hidden lay
the input data ݔ௜, ݅ ൌ 1,2, … , ܫ , are applied in the in
utput ܱ௞  in the output layer, which is compared to 
as been assigned in advance. If the error ݁௞ ൌ ݀௞ െ
245 
the 
the 
er j 
put 
the  
ܱ௞  
246 S. Omatu et al. 
 
 
Fig. 3. LVQ structures 
occurs, then the weighting coefficients ݓ௝௜,  ݓ௞௝  are corrected so that the error be-
comes smaller based on the error back-propagation algorithm. 
Next, we will show the LVQ: 
The structure of LVQ is two layered, consisting of an input layer and a competitive 
layer as shown in Fig. 3. 
In order to classify the odors we adopt a two-layered neural network based on the 
learning vector quantization method as shown in Fig. 3. Learning vector quantization 
is a supervised learning method for the purpose of pattern classification for input data. 
The learning method is given by the following steps: 
 
Step 1. Set the initial values of wij (j=1,2,…M, i=1,2,..,n) , T, and ߙ଴ (> 0) where 
T is the total iteration number for learning, n is the number of input, M is the number 
of neurons in cluster j, and ߙ଴ is the initial value of the learning rate. 
Step 2. First, calculate proximity to the coupling coefficient vector WJ of the input 
vector x and neuron j in the sense of Euclidean distance. The neuron with the closest 
coupling coefficient in the sense of Euclidean distance in the competitive layer neuron 
is detected by following equation for the input pattern.  
௝݀ ൌצ x െ ݓ௝ צൌ ඩ෍ሺݔ௜ െ ݓ௝௜ሻଶ
௡
௜ୀଵ
 
݀௖ ൌצ x െ ݓ௖ צൌ min௝ ௝݀      
Step 3. If the input vector and winning neuron c belong to the same class, then 
change ݓ௝ሺݐሻ by using the following equation: 
ݓ௝ሺݐ ൅ 1ሻ ൌ ݓ௝ሺݐሻ ൅ ߙሺݐሻ ቀݔ െ ݓ௝ሺݐሻቁ , ݆ ൌ ܿ 
                                                  ݓ௝ሺݐ ൅ 1ሻ ൌ ݓ௝, ݆ ് ܿ 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ଵܹ㻌
ݓଵଵ㻌
ݓଵଶ㻌
ݔଵ㻌 ݔଶ㻌 ݔ௡㻌
௡ܹ㻌
ݓଵ௡㻌
Cluster 1 Cluster 2 Cluster M 
... ... ... 
... 
Competitive 
layer 
Input layer 
 Multi-agent Technology to Perform Odor Classification 247 
where  
                               ߙሺݐሻ ൌ ߙ଴ ቀ1 െ ଵ்ቁ. 
If the input vector and neuron c belong to the different class, then 
change ݓ௝ሺݐሻ by using the following equation: 
ݓ௝ሺݐ ൅ 1ሻ ൌ ݓ௝ሺݐሻ െ ߙሺݐሻ ቀݔ െ ݓ௝ሺݐሻቁ , ݆ ൌ ܿ 
                                                  ݓ௝ሺݐ ൅ 1ሻ ൌ ݓ௝, ݆ ് ܿ. 
Step 4.  If t<T, Go to Step 2 
Using the above recursive procedure, we can train the odor data. 
2.3 Principle of QCM Sensors 
The QCM has been well-known to provide very sensitive mass-measuring devices in 
nanogram levels. Synthetic polymer-coated QCMs have been studied as sensors for 
various gas works as a chemical sensor. The QCM sensors are made by covering the 
surface with several kinds of a very thin membrane with about 1 mm, as shown in 
Fig. 4. The QCM sensor is integrated into a resonance circuit. If the film absorbs the 
odor molecules, the oscillation frequency is reduced since the mass of the vibrator is 
changed. 
Therefore, the frequency (of the QCM) will change according to the deviation of 
the weight due to the adsorbed odor molecular (odorant). In this paper we have used 
the materials shown in Table 1. The basic approach used here is a sol-gel method. The 
process is a wet-chemical technique used for the fabrication of both glassy and ceram-
ic materials. The manufacturing of a film was done by the following procedure. 
MTMS (1): Trimethylsilane, ethanol, water and nitric acid. (2) Once stirred, 2-
ethyl acrylate (PFOEA) was added to a solution of 1. 
Here, HTMS: Cܪଷ ௜ܵሺܱܥܪଷሻଷ and PFOEA: Fሺܥܨଶሻ଼Cܪଶܥܪଶ െ ܱܥܱܥܪ ൌ ܥܪଶ 
Table 1. Chemical materials used as the membrane.We used seven sensors using a solution of 
three types of in this paper. 
 
㼀㼍㼎㼘㼑㻌㻝㻌
㻿㼑㼚㼟㼛㼞㻌㼚㼡㼙㼎㼑㼞㻌㻹㼍㼠㼑㼞㼕㼍㼘㼟㻌㼛㼒㻌㼙㼑㼎㼞㼍㼚㼑
㼟㼑㼚㼟㼛㼞㻝 䃐
㼟㼑㼚㼟㼛㼞㻞 䃑
㼟㼑㼚㼟㼛㼞㻟 䀪
㼟㼑㼚㼟㼛㼞㻠 䃐
㼟㼑㼚㼟㼛㼞㻡 䀪
㼟㼑㼚㼟㼛㼞㻢 䃐
㼟㼑㼚㼟㼛㼞㻣 䃑
ߙ䠙ܥସܪଵଶ ଷܱܵ݅ሺͲǤͷ݃ሻ,ܥଵ଺ܪଵଽܨଵ଻ ଷܱܵ݅(0.015g),30%HN ଷܱ(10μL),ܪଶܱ ͲǤ͵Ͳ݈݉ ǡ ݁ݐܽ݊݋݈ሺ͵ǤͲ݈݉ሻ
ߚ䠙ܥସܪଵଶ ଷܱܵ݅ሺͲǤͷ݃ሻ,ܥଵ଺ܪଵଽܨଵ଻ ଷܱܵ݅(0.030g),30%HN ଷܱ(10μL),ܪଶܱ ͲǤͳͷ݈݉ ǡ ݁ݐܽ݊݋݈ሺ͵ǤͲ݈݉ሻ
ᢗ䠙ܥସܪଵଶ ଷܱܵ݅ሺͲǤͷ݃ሻ,30%HN ଷܱ(10μL),ܪଶܱ ͲǤ͵݈݉ ǡ ݁ݐܽ݊݋݈ሺ͵Ǥʹ݈݉ሻ
248 S. Omatu et al. 
The QCM sensor is integ
sor absorbs odorant, the we
is also reduced when the m
the crystal oscillation will b
2.4 Odor Sensing Hard
Generally, this type of syst
signed to detect some speci
teristics in response to diffe
the ability to detect the od
nose (EN) systems), shown
human olfactory system. O
Fig. 6. 
Permeater can continuou
organic gas) for a long tim
which generates a continuo
QCM sensors, listed in Tab
The odors used here are sho
First, allow the dry air t
sampling box. Then, the re
counter as data. In addition
verts the data stored in the f
 
Fig. 4. Principle of QCM sensor 
rated into a resonance circuit. If the membrane of the s
ight of quartz plate will be modified. Oscillation freque
embrane absorbs odorants. Thus, the original frequency
ecome smaller according to the density of odorants 
ware System 
em, for example an air purifier, a breathalyzer, etc., is 
fic odor. Each of the QCM membranes has its own char
rent odors. When combining many QCM sensors togeth
or is increased. Odor sensing systems (called electro
 in Fig. 5, were developed based on the concept of 
dor generating machines are using Permeater, as shown
sly generate standard gas of many kinds (inorganic 
e. The process takes place in the standard gas genera
us standard of trace gas concentration. The combination
le 1, is used as the olfactory receptors in the human no
wn in Table 2. 
  
Fig. 5. Odor sensing systems 
o flow so the odor coming from the Permeater get to 
action of the QCM sensor is accumulated into freque
, the gas is exhausted from the sampling box. The PC c
requency counter into a text file. 
en-
ncy 
 of 
de-
ac-
er, 
nic 
the 
 in 
and 
tor, 
 of 
se. 
the 
ncy 
on-
 Multi-agent Technology to Perform Odor Classification 249 
    
Fig. 6. Permeater and QCM sensors 
Table 2. Kinds of odors measured in this experimet 
 
2.5 Measurement of Odor Data 
We have measured four types of odors as shown in Table 2. The sampling frequencies 
are 1[Hz]. Diffusion tubes are used to control the density of gases. This is because it 
is possible to generate the gas at various concentrations by using a diffusion tube 
through Permeater. Odor data are measured for 900 [s]. They may include impulsive 
noises due to the typical phenomena of QCM sensors. To remove these impulsive 
noises we adopt a median filter which replaces a value at a specific time by a median 
value among neighboring data around the specific time. In  Fig. 7 we show the mea-
surement data for the symbol A where the horizontal axis is the measurement time 
and the vertical axis is the frequency deviation from the standard value (20M[Hz]) 
after passing through a three-point median filter. 
  
Fig. 7. Measurement of odor data 
p
Symbols Kind of odors
A ethanol
B toluen
C ethly acetatate
 
250 S. Omatu et al. 
Here, seven sensors are used. The maximum value for each sensor among seven 
sensors is selected as a feature value for the sensor. Therefore, for one odor, there are 
seven sensor values, which will be used for classification. 
3 Conclusions and Results 
In order to classify the feature vector by using error-back propagation, we allocate the 
desired output for the input feature vector, which is a seven-dimensional vector, as 
shown in Table 3. By adding the coefficient of variation to the usual feature vector, 
the variations for odors are reduced. The training was performed until the total error 
was less than or equal to 1×10−2 where η=0.8. 
Table 3. Training data set for ethanol (A), toluene (B), and ethyl acetate (C) 
Symbols Output A Output B Output C 
    A 1 0 0 
    B 0 1 0 
    C 0 0 1 
 
We have examined two algorithms, a learning vector quantization, and error back 
propagation. In learning vector quantization and error back propagation, the training 
sample number P’= 8 and test sample number is three. 
The total number of classification of 100 test samples is checked. The results are 
summarized in Table 4 and Table 5. 
Table 4. Classification results for learning vector classification (LVQ) 
Odor data Classification results (97%) 
A B C Correct 
A 100 0 0 100 
B 1 97 2 97 
C 4 1 95 95 
Table 5. Classification results for layered neural networks 
Odor data Classification results (88%) 
A B C Correct 
A 86 14 0 86 
B 14 84 2 84 
C 2 3 95 95 
 
We have presented the reliability of a new EN system designed from various kinds 
of QCM sensors. We have shown that after training the neural network for each odor, 
we were able to classify the original odor from the mixed odors in the case of two 
odors.  
 Multi-agent Technology to Perform Odor Classification 251 
In addition, we have proposed and developed a multi-agent system which is able to 
increment the percentage of correct outputs and reduce the training time by sharing all 
the data between other similar odor detecting systems, and correcting the error be-
tween their sensors. The multi-agent system implemented in PANGEA performs 
communication, control and data management services in a distributed and flexible 
way. In the case study presented, a classification method is implemented. However 
the use of PANGEA makes it possible to extend the system by using new methods for 
the classification of odors and making the system scalable in terms of functionality. 
Acknowledgements. This work has been carried out by the project Sociedades Hu-
mano-Agente: Inmersión, Adaptación y Simulación. TIN2012-36586-C03-03. Minis-
terio de Economía y Competitividad (Spain). Project co-financed with FEDER funds. 
References 
[1] Want, R., Hopper, A., Falcao, V., Gibbons, J.: The Active Badge Location System. ACM 
Transactions on Information Systems 10(1), 91–102 (1992) 
[2] Lim, C.H., Anthony, P., Fan, L.C.: Applying multi-agent system in a context aware. Bor-
neo Science 24, 53–64 (2009) 
[3] Ahmed, A., Ali, J., Raza, A., Abbas, G.: Wired Vs Wireless Deployment Support For 
Wire-less Sensor Networks. In: 2006 IEEE Region 10 Conference TENCON 2006, pp. 1–
3 (2006) 
[4] Lee, M., Yoe, H.: Comparative Analysis and Design of Wired and Wireless Integrated 
Networks for Wireless Sensor Networks. In: Software Engineering Research, Manage-
ment & Applications, SERA 2007, pp. 518–522 (2007) 
[5] Bajo, J., Corchado, J.M., De Paz, Y., De Paz, J.F., Rodríguez, S., Martin, Q., Abraham, 
A.: SHOMAS: Intelligent Guidance and Suggestions in Shopping Centres. Applied Soft 
Com-puting. Elsevier Science 9(2), 851–862 (2009) 
[6] Fonseca, S.P., Martin, P., Griss, L., Letsinger, R.: An Agent Mediated E-Commerce Envi-
ronment for the Mobile Shopper, Hewlett-Packard Laboratories, Technical Report, HPL-
2001-157 (2001) 
[7] Kowalczyk, R., Ulieru, M., Unland, R.: Integrating Mobile and Intelligent Agents in Ad-
vanced E-commerce: A Survey. In: Kowalczyk, R., Müller, J.P., Tianfield, H., Unland, R. 
(eds.) NODe-WS 2002. LNCS (LNAI), vol. 2592, pp. 295–313. Springer, Heidelberg 
(2003) 
[8] Carabelea, C., Boissier, O.: Multi-Agent Platforms on Smart Devices: Dream or Reality? 
In: Proceedings of the SOC 2003, pp. 126–129 (2003) 
[9] Fok, C., Roman, G., Lu, C.: Agilla, A Mobile Agent Middleware for Self-Adaptive Wire-
less Sensor Networks. ACM Transactions on Autonomous and Adaptive Sys. (3), 1–26 
(2009) 
[10] Meyer, G.G., Främling, K., Holmström, J.: Intelligent Products: A survey. Computers in 
Industry 60(3), 137–148 (2009) 
[11] Kazanavicius, E., Kazanavicius, V., Ostaseviciute, L.: Agent-based framework for em-
bedded systems development in smart environments. In: Proceedings of the 15th Interna-
tional Conference on Information and Software Technologies IT, pp. 194–200 (2009) 
[12] Tynam, R., Ruzzelli, A., O’Hare, G.M.P.: A Methodology for the Deployment of Multi-
Agent Systems on Wireless Sensor Networks. Mutiagent and Grid Systems, 491–503 
(2006) 
[13] Rogers, A., Jennings, N.R., Corkill, D.: Agent Technologies for Sensor Networks. IEEE 
Intelligent Systems 24(2), 13–17 (2009) 
252 S. Omatu et al. 
[14] Griol, D., García-Herrero, J., Molina, J.M.: Combining heterogeneous inputs for the de-
vel-opment of adaptive and multimodal interaction systems. Advances in Distributed 
Computing And Artificial Intelligence Journal (2013) ISSN: 2255-2863 
[15] Tapia, D.I., Abraham, A., Corchado, J.M., Alonso, R.S.: Agents and ambient intelligence: 
case studies. Journal of Ambient Intelligence and Humanized Computing 1(2), 85–93 (2010) 
[16] Bajo, J., Corchado, J.M.: Evaluation and monitoring of the air-sea interaction using a 
CBR-Agents approach. Case-Based Reasoning Research and Development 50 (2005) 
[17] Corchado, J.M., De Paz, J.F., Rodríguez, S., Bajo, J.: Model of experts for decision support 
in the diagnosis of leukemia patients. Artificial Intelligence in Med. 46(3), 179–200 (2009) 
[18] De Paz, J.F., Rodríguez, S., Bajo, J., Corchado, J.M.: Case-based reasoning as a decision 
support system for cancer diagnosis: A case study. International Journal of Hybrid Intelli-
gent Systems 6(2), 97–110 (2009) 
[19] Tapia, D.I., Rodríguez, S., Bajo, J., Corchado, J.M.: FUSION@, a SOA-based multi-agent 
archi-tecture. In: International Symposium on Distributed Computing and Artificial Intel-
ligence 2008 (DCAI 2008), pp. 99–107 (2008) 
[20] Corchado, J.M., Aiken, J.: Hybrid artificial intelligence methods in oceanographic fore-
cast models. IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications 
and Reviews 32(4), 307–313 (2002) 
[21] Bajo, J., De Paz, J.F., Rodríguez, S., González, A.: Multi-agent system to monitor oceanic 
envi-ronments. Integrated Computer-Aided Engineering 17(2), 131–144 (2010) 
[22] Hardy, A., Bouhafs, F., Merabti, M.: A Survey of Communication and Sensing for Energy 
Management of Appliances. International Journal of Advanced Engineering Sciences and 
Technologies 3(2), 61–77 (2010) 
[23] Zato, C., Villarrubia, G., Sánchez, A., Bajo, J., Corchado, J.M.: PANGEA: A New Plat-
form for Developing Virtual Organizations of Agents. International Journal of Artificial 
Intelligence 11(A13), 93–102 (2013) 
[24] Omatu, S., Yano, M.: Intelligent Electronic Nose System Independent on Odor Concen-
tration. In: Abraham, A., Corchado, J.M., González, S.R., De Paz Santana, J.F. (eds.) In-
ternational Symposium on DCAI. AISC 91, pp. 1–9. Springer, Heidelberg (2011) 
[25] Ferber, J., Gutknecht, O., Michel, F.: From Agents to Organizations: an Organizational 
View of Multi-Agent Systems. In: Giorgini, P., Muller, J., Odell, J. (eds.) AOSE 2003. 
LNCS, vol. 2935, pp. 214–230. Springer, Heidelberg (2004) 
[26] Rodríguez, S., de Paz, Y., Bajo, J., Corchado, J.M.: Social-based planning model for mul-
tiagent systems. Expert Systems with Applications 38(10), 13005–13023 (2011) 
[27] Foster, I., Kesselman, C., Tuecke, S.: The anatomy of the grid: Enabling scalable virtual 
organizations. Int. J. High Perform. Comput. Appl. 15(3), 200–222 (2001) 
[28] Pavon, J., Sansores, C., Gomez-Sanz, J.J.: Modelling and simulation of social systems 
with INGENIAS. International Journal of Agent-Oriented Software Engineering 2(2), 
196–221 (2008) 
[29] Garijo, F., Gómes-Sanz, J.J., Pavón, J., Massonet, P.: Multi-agent system organization: 
An engineering perspective. In: Pre-Proceeding of the 10th European Workshop on Mod-
eling Autonomous Agents in a Multi-Agent World, MAAMAW (2001) 
[30] Milke, J.A.: Application of Neural Networks for discriminating Fire Detectors. In:10 th 
International Conference on Automatic Fire Detection, AUBE 1995, Duisburg, Germany, 
pp. 213–222 (1995) 
[31] Charumporn, B., Yoshioka, M., Fujinaka, T., Omatu, S.: Identify household burning smell 
using an electronic nose system with artificial neural networks. In: Proceedings of the 
IEEE International Symposium on Computational Intelligence in Robotics and Automa-
tion, vol. 3, pp. 1070–1074. IEEE (2003) 
[32] Fujinaka, T., Yoshioka, M., Omatu, S., Kosaka, T.: Intelligent Elec-tronic Nose Systems 
for Fire Detection Systems Based on Neural Networks. In: The Second International Con-
ference on Advanced Engineering Computing and Applications in Sciences, ADVCOMP 
2008, pp. 73–76. IEEE (2008)