Information Technology Reference
In-Depth Information
some plans that satisfy the goals are collected. In the fourth and last step, the best plan
is chosen and is converted in a composite web service, encoded in BPEL.
A system which was developed recently and is analyzed in [6] is the system
PORSCE. The approach is based on transforming the web services composition prob-
lem to a planning problem. The straight forward mapping of these two fields is
exploited and the OWL-S descriptions of the available web services are used to con-
struct PDDL plan files. The initial state is derived by the data given as input to the
final web service by the user, whereas the goals are reflected by the desired outputs.
The operators of the problem correspond to the available atomic web services that can
be used. Their preconditions are mapped to the inputs of the services and theirs results
to the outputs. Simultaneously, the ontologies that are connected to the types of the
parameters of the available web services are used so as for the semantics of the
concepts to be provided. The system starts by representing the composition problem
with planning terms. Then, a solution to the problem is provided by an external plan-
ner, such as LPG-td [3], [4] or JPlan [8], according to the user's selection. Finally, the
quality of the produced plan is measured based on some quality measures selected by
the user at the beginning of the process and the results are provided to the user. There
is also the possibility of replacing instantly some of the web services in the plan with
other relevant, as they are discovered during the planning process.
Another approach that exploits the similarities between the AI planning and
semantic web services composition research fields is the OWLS-Xplan [9]. This sys-
tem uses the OWL-S descriptions of the available web services, the relevant OWL
ontologies that define the types of the parameters in the descriptions and a planning
query as input. After some preprocessing of the above data and the execution of the
Xplan planning algorithm, the result is a plan describing the sequence of composed
services that satisfies the goals.
The OWLS-Xplan approach consists of two basic parts. The first one is an
OWLS2PDDL converter which converts the OWL-S descriptions along with the
OWL ontologies to the equivalent PDDL domain and problem of the composition.
Specifically, the conversion results to descriptions of the domain and problem in a
XML dialect of PDDL (developed by the authors), referred to as PDDXML, that
simplifies parsing, reading and communicating the descriptions using SOAP. An
atomic operator is directly related to a service profile as they both provide a general
description of their instances, actions and web services, respectively. A complex
action can be linked to a service model that describes how simpler actions should
cooperate to result to the composite one. Finally, the methods used in HTN planning
are related to composite web services and may be used by the planner as a hierarchic-
al task network during the planning process.
The second part of OWLS-Xplan is the developed heuristic hybrid Xplan AI
planner that combines the benefits of the action-based FF-planner [7] with HTN plan-
ning. Xplan always finds a solution, if it exists in the state space, over the space of
possible plans, in contrast to HTN approaches. It combines guided local search with
graph planning and a simple form of hierarchical task networks to produce a plan.
Also, a re-planning component is included to improve flexibility is cases changes
happen in the world during planning, a property well needed in semantic web services
composition field.
Search WWH ::




Custom Search