J. Software Engineering & Applications, 2011, 4, 585-589
doi:10.4236/jsea.2011.410068 Published Online October 2011 (http://www.SciRP.org/journal/jsea)
585
Architecture of Dynamic Service Composition
Using Logic Petri Nets
Xia Jiang1,3, Bing Wu2, Yuyue Du3
1Laiwu Vocational and Technical College, Laiwu, China; 2Information Centre of Shandong Province, Jinan, China; 3College of In-
formation Science and Engineering, Shandong University of Science and Technology, Qingdao, China.
Received August 8th, 2011; revised September 2nd, 2011; accepted September 11th, 2011.
ABSTRACT
Services are preprocessed based on clustering techniques in the levels of function and similarity of services. Service
clustering is modeled by means of the passing value indeterminacy of logic Petri nets. Thus service clustering can be
described, and the services with indeterminacy parameters can be modeled formally. Thus service composition is inves-
tigated under the classified architecture of service clustering. The architecture of dynamic service composition is given,
and logic Petri nets are used for service realization. Services can be combined rapidly and automatically, and the path
of a composition service can be constructed by automatic inference.
Keywords: Service Clustering, Service Composition, Architecture, Modeling, Petri net
1. Introduction
In the process of promoting global economic integration
and informatization, Information industry patterns are
changed by the integration and development and of in-
formation technology, and Web services has become one
of the hot issues in current researches. The relevant re-
searches of services mainly focus on service description,
service discovery, service matching and service composi-
tion and so on. Web services mainly are described by the
existing service description languages as follows. RDF
(Resource Description Framework) is used to describe
the metadata and the relationships between them. The
semantic description ability of OWL (Ontology Web
Language) is stronger than RDFS, and OWL is suitable
for the machine automatic reasoning. OWL-S (Ontology
Web Language for Services) is specially used to describe
the semantics of Web services, then the Web services can
be comprehensible by machines, and it becomes possible
to discover, execute and composite Web services auto-
matically.
There are many uncertain factors of the information
services under the network environment such as the un-
certainties of user requirements, collaborative services
and aggregate service semantic. For instance, the number
and description of the keywords input by users may be
uncertain in Google search engine. Then different search
results should be given by Google search engine accord-
ing to the different inputs and logical relationships of the
keywords. However, specific user requirements and sin-
gle service description mainly are considered in the ex-
isting researches of service semantic, and the modeling
and analysis of specific problems are completed based on
requirement definitions, service searching and composi-
tion. Then the service applications are realized by the
way of resource aggregation and synergy. Nevertheless,
there is no a set of normative model theory to represent
requirement uncertainty. Then the above uncertainty fac-
tors in service composition are difficult to deal with and
the qualities of services are hard to guarantee. Also, it is
difficult to meet dynamic and various user requirements.
Therefore, a set of effective model theory is established
in view of the existing uncertain factors in service com-
position. According to the dynamic and various user re-
quirements under the network environment, a design
method of correct service process logic is studied to adapt
dynamic environment. And it has become the current
challenges of information services.
In the service classification systems, the services at the
bottom have the same or similar functions. These services
have similar concepts and semantic information, and they
are regarded as a service cluster though the parameter
number or parameter names of them are not exactly the
same. With the popularization of Web service applica-
tions, Web service quantity has increased dramatically.
So many studies improve service searching efficiency by

Architecture of Dynamic Service Composition Using Logic Petri Nets
586
use of clustering [1-4].
A Logic Petri Net (LPN) is high-level abstraction of
Petri nets with inhibitor arcs (IPN), and its expression
ability is equivalent to IPN in expression ability [5]. LPN
sorts all indeterminacy into the input and output indeter-
minacy, and describes them respectively by logic input
and output transitions. In fact, because LPN can effect-
tively describe the property of passing value indetermi-
nacy and batch processing function of systems, it has
successfully been used in the modeling and analysis of
stock trading systems, e-commerce systems and real-time
cooperative systems [6-9]. In LPN, the logic expressions
of s can effectively indicate the uncertain parameters of
service clusters. In this paper, a service cluster is de-
scribed as a LPN element (a net element includes the
logic transition and its input/output places) where the
input/output places denote cluster input/output parame-
ters and the logic expression denotes the uncertain rela-
tionships of them. Therefore, the whole service classify-
cation system is constructed based on the net elements of
LPN. In addition, service clusters are considered as the
minimum elements, then the efficiency and accuracy of
service composition can be improved by using reasoning
and calculation theory of LPN.
2. Logic Petri Nets
For more than 40 years, Petri nets theory constantly de-
velops and its description ability continuously strength-
ens. Logic Petri nets (LPN) [10-13] are proposed in order
to improve the modeling capability of Petri nets. In fact,
LPN can describe and analyze the property of passing
value indeterminacy and batch processing function, and it
can efficiently mitigate the problem of state explosion to
some extent. Some basic definitions are given as follows.
Definition 1. If LN = (P, T, F, DT, I, O), LPN = (LN,
M) is a Logic Petri net if and only if
1) P is a finite set of places;
2) T = TD  TI  TO is a finite set of transitions, T  P
 , t  TI  TO: t  t =  , and P, TD, TI, TO are
disjoint with each other, where:
a) TD denotes a set of delay transitions, and the delay
time is described as τ;
b) TI denotes a set of logic input transitions, and t 
TI, all input places of t are restricted by a logic input ex-
pression fI;
c) TO denotes a set of logic output transitions, and t 
TO, all output places of t are restricted by a logic output
expression fO;
3) F  (P  T)  (T  P) is a finite set of arcs;
4) DT is a real function, t  TD: DT(t) = τ  R, τ is the
delay time;
5) I is a mapping from a logic input transition to a
logic input expression, i.e. t  TI, I(t) = fI(t);
(6) O is a mapping from a logic output transition to a
logic output expression, i.e. t  TO, O(t) = fO(t);
(7) M: P  {0,1} is a marking function, where p  P,
M(p) is the number of tokens in p;
(8) Transition firing rules:
a) t  TD: DT(t) = τ, if p  •t: M(p) = 1, t is enabled
at M; if the enabled time of t is equal to or greater than τ,
then t can fire, and generates a new state M′, p•t-t•:
M′(p) = M(p) − 1, pt•-•t: M′(p) = M(p) + 1, otherwise,
M′(p) = M(p);
b) t  TI: I(t) = fI, if fI|M = T, in other words, •t sat-
isfy a logic input expression fI at M, then t is enabled at
M; if t is enabled, then it can fire, and generates a new
state M′, p  •t: M′(p) = 0, pt•: M′(p) = M (p) + 1,
p  •t  t•: M′(p) = M(p);
c) t  TO: O(t) = fO, if p  •t: M(p) = 1, then t is
enabled an M, if t is enabled, then it can fire, and gener-
ates a new state M′, p  •t: M′(p) = M(p)-1, p  t• 
•t: M′(p) = M(p). while t• should satisfy the expression
fO|M′ = T, in other words, t• must satisfy a logic expres-
sion fO at M′.
In LPNs, the graphical representation of logic input/
output transitions is shown in Figure 1.
3. Service Clustering and LPN
Representation
A unified classification mechanism is provided by UDDI
(Universal Description, Discovery, and Integration) for
the users. However, these classification standards are
relatively coarse and not unified. The services are divided
mostly based on service fields or regional and service
function cannot be correctly reflected under these classi-
fication methods. In addition, the services have rich se-
mantics, but the matching semantic processing ability can
not be provided by UDDI, so the effective service or-
ganization and management are unable to be realized
only relay on the classification methods provided by
UDDI. In this paper, the similarity judgments of service
function and behavior are added based on the UDDI. If
the similarities of the services are within the certain scope,
then put them into the same service cluster.
Figure 1. Representation of Logic input/output transitions
in LPN.

Architecture of Dynamic Service Composition Using Logic Petri Nets587
The clustering methods are introduced in order to re-
search dynamic service composition based on service
clusters. In service classification systems, the services at
the bottom have the same or similar functions. These
services have similar concepts and semantic information,
and they are regarded as a service cluster though the pa-
rameter number or parameter names of them are not ex-
actly the same.
Service function refers to what is a service used to do
and what effect dose a service realize. At present, the
service function is judged by two parts: one is the in-
put/output parameters; the other is the internal structure
of a service. Part of the input/output is service and the
other part is the internal structure of service. In order to
judge whether two services are contained in the same
cluster, service encapsulation is a common method and
only the service input/output parameters should be con-
sidered. There have been many papers on input/output
research, and the corresponding determination methods
have been given. For example, the method of in-
put/output similarity judgment proposed by Sun Ping [4]
is as follows: First, the input/output parameters of two
services are paired off respectively and the relationships
between two parameter sets are established; then, the
ontology distance between each pair of parameters should
be calculated and the average ontology distance of pa-
rameter pairs can be regarded as the input/output similar-
ity of two services. For example, suppose that the input
parameters of services wsj, wsk respectively are Inputj,
Inputk, and Inputj = {ij1, ij2, ···}, Inputk = {ik1, ik2, ···}. The
average ontology distance of parameter pairs is denoted
as Dis(Inputj, Inputk) =

ja ka
1
1dis i,i
l
a
l
.
According to the definition of LPN, a service cluster
can be described as a LPN element where the input/out-
put places denote input/output parameters and the logic
expression denotes the uncertain relationships of them.
Therefore, the whole service classification system is con-
structed based on the net elements of LPN. In addition,
service clusters are considered as the minimum elements,
then the efficiency and accuracy of service composition
can be improved by using reasoning and calculation the-
ory of LPN. The relationship between the key input/out-
put parameters is “and”, while the relationship between
the optional input/output parameters is “or”. Therefore,
the one-to-one mappings from service clusters to LPN
elements are established and the services in the library
are mapped into LPN elements. For example, there are
three atomic services S1, S2, S3 in the cluster of ticket
reservation query. InputS1 = {Departure City, Departure
Time, Ticket Price}, InputS2 = {Departure City, Aircraft
Type, Departure Time, Ticket Price}, InputS3 = {Depar-
ture City, Subordinate Airline, Departure Time, Ticket
Price}. OutputS1 = OutputS2 = OutputS3 = {Subordinate
Airline, Aircraft Type, Flight Number, DepartureTime,
Arrival Time, Ticket Price}. The encapsulation forms of
the three atomic services are shown in Figure 2, and the
corresponding LPN element is shown in Figure 3.
The input parameters of the atomic services in the ser-
vice cluster are integrated to form the logic input expres-
sion fI. The necessary input parameters are called key
parameters. In Figure 2, V1, V2 andV3 are key parameters
in logic input expression fI. In the example above, the
atomic services have different input but the same output
parameters. Likewise, when the atomic services have
different output parameters, the logic output expression
can be used to denote the output of the service cluster.
4. Dynamic Service Composition Based on
Service Clusters
The architecture of dynamic service composition is pro-
posed in this section and it is shown in Figure 4. The
Figure 2. Three ticket reservation query services with dif-
ferent input parameters.
Figure 3. The corresponding LPN element of ticket reser-
vation query service cluster.

Architecture of Dynamic Service Composition Using Logic Petri Nets
588
server needs to search the service clusters to match the
user request, and service cluster searching also concerns
input/output, but it only needs to match the key parame-
ters of the logic input/output expression. Therefore, the
matching complexity is reduced. And a set of service
clusters to achieve the user requests are obtained after
service cluster searching. Since the services with the same
or similar functions are abstracted as a service cluster, the
returned service cluster set contains some service clusters
which can realize some function during the service flow
and the function of them are not the same with each other.
Once the availability of these service clusters and the
logical relationship between them are confirmed, the ab-
stract service flow is completed.
Normally, the logical relationships between service
clusters can be determined according to the flow of prac-
tical things. The service clusters are denoted by LPN
elements, and then the redundant model of service flow
can be constructed based on the links rules of LPN ele-
ments. The atomic services to realize user requests are
contained in this redundant model by the logical forms.
Multiple service flows which can achieve the user goals
should be found out from the redundant model, and then
the optimal service flow can be obtained by the compare-
son among the flows. Finally, the optimal service flow
can be executed.
The framework of dynamic service composition takes
service clusters as search granularity and LPN as techni-
cal support. Because LPN is proposed to solve the ex-
pression problem of the model with uncertain parameters,
LPN is suitable to model service clusters. In addition, the
introduction of logical expressions helps to complete the
complex LPN reasoning, and it can solve the problems of
model hugeness and reasoning chaos.
Figure 5 shows the technologies used in every part of
the framework and the logical relationships between them.
The framework is constructed based on clustering tech-
nology and existing LPN theory researches, and the re-
lated theory of LPN is used in service composition to
realize the dynamic construction of redundant service
model and path selection. In foundation layer, the func-
tion similarities are judged based on UDDI field classify-
cation, service clustering and the service classification
methods are researched, and the LPN representation of
Request Service cluster
searching
Service
cluster
database
Library of
service clusters
meeting request
Matching links of
service clusters
Reasoning and
validation of
Service path
Coarse granularity index
mechanism of service
cluster net elements
Redundant
service
original model
Path selectionExecution
Multiple
service flows
Validation
rules of net
element links
Reasoning way
combining Petri nets
and logical expression
Comparison of
non-function
parameters
Response
Figure 4. Dynamic service composition based on service clusters.
Multi-level service classification mechanism with semantic association
Logic transition
representations of
service clusters
Demand
related service
clusters
Representation in
disjunctive normal forms
Reasoning analysis
technique of LPN
Quotient space
algebra
Ontologies
Logic Petri
nets
UDDI
Dynamic linking technology
of LPN elements
Modeling
technology of LPN
User
request
Matching key
parameters
Evolution of
logical transitions
Construction of redundant
service model
Multiple service
executive flows
Service cluster
searching
Service clusters
matching links
Service path reasoning
and validation
Path
selection
and
execution
Service clustering
and service cluster
classification
LPN representation
of service clusters
Coarse granularity index
mechanism based on net
elements with semantics
Figure 5. The technology roadmap of framework implementation.

Architecture of Dynamic Service Composition Using Logic Petri Nets589
service clusters is given. In application research layer, the
dynamic link and reasoning analysis technology of LPN
are used in the service flow reasoning to obtain multiple
service paths.
5. Conclusions
A service composition framework is proposed in this
paper. It takes service clusters as the minimum granular-
ity, then the scale of service model is greatly reduced and
service searching efficiency is improved. Logical rea-
soning technology and process evolution method both are
used to realize the service composition, then the problem
of state explosion can be efficiently mitigated. The re-
searches of this framework will be significant for service
composition because the automatic service composition,
estimate and quality control algorithm can be effectively
completed under this framework though there are large-
scale services.
A modeling method of service clusters is proposed in
this paper and the LPN representation of service clusters
is completed. The future researches mainly focus on the
following aspects: Doing researching on the multi-layer
classification indexing mechanism of service clusters to
improve the searching efficiency of service cluster; im-
proving the reasoning analysis technology of LPN to ob-
tain the specific reasoning process of service paths.
6. Acknowledgements
This work was supported in part by the National Natural
Science Foundation of China under Grants 61170078 and
60773034; the National Basic Research Program of China
(973 Program) under Grant 2010CB328101; and the Sci-
entific and Technological Developing Program of Shan-
dong Province of China under Grant 2011GGX10114.
REFERENCES
[1] S. Rajagopal and S. T. Selvi, “Semantic Grid Service Dis-
covery Approach Using Clustering of Service Ontolo-
gies,” Proceedings of IEEE TENCON, Hong Kong, 14-17
November 2006, pp. 1-4.
[2] S. Ram, Y. Hwang and H. Zhao, “A Clustering Based Ap-
proach for Facilitating Semantic Web Service Discovery,”
Proceedings of the 15th Annual Workshop on Information
Technologies & Systems, Las Vegas, 10-11 March 2006,
pp. 1-6.
[3] R. Nayak and B. Lee, “Web Service Discovery with Addi-
tional Semantics and Clustering,” Proceedings of the IEEE/
WIC/ACM International Conference on Web Intelligence,
Washington DC, 2-5 November 2007, pp. 555-558.
[4] P. Sun and C. Jiang, “Using Service Clustering to Facili-
tate Process-Oriented Semantic Web Service Discovery,”
Chinese Journal of Computers, Vol. 31, No. 8, 2008, pp.
1340-1353.
[5] Y. Y. Du and B. Q. Guo, “Logic Petri Nets and Equiva-
lency,” Information Technology Journal, Vol. 8, No. 1,
2009, pp. 95-100. doi:10.3923/itj.2009.95.100
[6] Y. Y. Du and C. J. Jiang, “Formal Representation and
Analysis of Batch Stock Trading Systems by Logical Petri
Net Workflows,” Lecture Notes in Computer Science, Vol.
2495, 2002, pp. 221-225.
[7] Y. Y. Du and C. J. Jiang, “Towards a Workflow Model of
Real-Time Cooperative Systems,” Lecture Notes in Com-
puter Science, Vol. 2885, 2003, pp. 452-470.
[8] Y. Y. Du, C. J. Jiang and M. C. Zhou, “Modeling and
Analysis of Real-Time Cooperative Systems Using Petri
Nets,” IEEE Transactions on Systems, Man, and Cyber-
netics, Part A: Systems and Humans, Vol. 37, No. 5, 2007,
pp. 643-654. doi:10.1109/TSMCA.2007.902622
[9] W. Liu and Y. Y. Du, “Modeling Multimedia Synchroni-
zation Using Petri Nets,” Information Technology Journal,
Vol. 8, No. 7, 2009, pp. 1054-1058.
doi:10.3923/itj.2009.1054.1058
[10] Y. Y. Du and C. J. Jiang, “On the Design and Temporal
Petri Net Verification of Grid Commerce Architecture,”
Chinese Journal of Electronics, Vol. 17, No. 2, 2008, pp.
247-251.
[11] W. Luo, X. Qin, X.-C. Tan, K. Qin and A. Manzanares,
“Exploiting Redundancies to Enhance Schedulability in
Fault-Tolerant and Real-Time Distributed Systems,” IEEE
Transactions on Systems, Man, and Cybernetics, Part A:
Systems and Humans, Vol. 39, No. 3, 2009, pp. 626-639.
doi:10.1109/TSMCA.2009.2013192
[12] A. L. Feller, T. Wu, D. L. Shunk and J. Fowler. “Petri Net
Translation Patterns for the Analysis of Ebusiness Col-
laboration Messaging Protocols,” IEEE Transactions on
Systems, Man, and Cybernetics, Part A: Systems and Hu-
mans, Vol. 39, No. 5, 2009, pp. 1022-1034.
doi:10.1109/TSMCA.2009.2025022
[13] D. X. Xia, X. Y. Xie and Y. Xu, “With an Application to
Workflow Management Based on Colored Petri Net,”
Proceedings of 3rd International Conference on Anti-
counterfeiting, Security, and Identification in Communi-
cation, Hong Kong, 20-22 August, 2009, pp. 596-599.