![](bg2.png)
road map that guides research in this area (PLANET,
2003) and aims at integration between workflow
process modeling and planning tools.
Workflow tools are often used in companies,
which generally have uncertain and changing
environments. These changing environments are the
result of marked demands such as client exigencies,
competition or new governmental laws. In order to
satisfy these demands, two types of tools are
required. The first type of tool should be used to
model, to modify, to simulate and to optimize
automatically existing processes in agreement with
the business rules of the company. The second type
of tool should be used to execute, monitor and adapt
dynamically processes resulting from the
modifications generated by the first tool. These two
types of tools constitute the main structure of a
workflow application. In this article, emphasis is
given to the former tool. We claim that techniques of
AI Planning can increase the efficiency of process
modeling and produce sound models.
This article is structured as follows. In section 2, we
describe the main concepts related to workflow
modeling. Section 3 presents AI planning concepts.
Section 4 shows the architecture in which our
proposal of planning and workflow integration was
built. The work developed is presented and detailed
in sections 5 and 6. Finally, section 7 discusses
related research and section 8 draws the conclusions
of this article.
2 WORKFLOW MODELLING
We can analyze a workflow system by taking into
account its various functions which cover three main
areas: process modeling, execution, and
management. In this article, emphasis is given to
process modeling.
The modeling of a process begins by translating
features from the real world to the computational
environment, generating a formal model by means
of appropriate modeling techniques. A model must
contain all the important information about the
process, such as: start and finish conditions,
execution rules, users or agents, information or
documents to be manipulated, interaction with
external applications etc. Below we describe the
main components and features of a modeling task.
Activities: The main element in a workflow is an
activity which must be completed for the conclusion
of the process goal to take place. To execute an
activity an actor must assume a role related to this
activity. This actor can be a person, an application,
or a computerized agent.
Chaining of Activities: The activities in a
workflow can be organized in three different ways:
sequentially, in parallel or conditionally.
Sequentially means that as soon as one activity is
executed, the following is activated. The subsequent
activity cannot be initiated until the current one is
concluded. Activities executing in parallel are
activated simultaneously. They do not need to be
concluded at the same time, because they can follow
different criteria or they can demand distinct
operations.
Eventually, these parallel flows either converge and
become a sequential flow or reach the end of the
process. The conditional chaining appears when the
next activity to be executed is based on a decision.
This decision will be made based on information
contained in the process. Rules for conditional
chaining must be created during the process
modeling.
3 AI PLANNING
AI planning seeks to determine an orderly set of
actions that, when executed by one or more agents
from an initial state (that satisfies the start
conditions), results in a final state, which satisfies a
goal. The sequence of actions makes up a plan
(Russell and Norvig, 2003). The PLANET roadmap
subtly complements this definition by introducing
the process concept. A process is a description of an
ordered set of activities. A plan is a description of a
sequence of actions that lead to the fulfillment of an
objective, i.e., an instantiated process (PLANET,
2003). These definitions are complementary, and
they introduce the process as a set of all the valid
plans.
We have used the classical planning model as a
starting point for applying planning techniques to the
workflow domain.
The evolution of planning techniques has also
promoted the evolution of its representation by using
an appropriate language. Important milestones in
this evolution are STRIPS, the ADL language, and
more recently PDDL (Russell and Norvig, 2003).
Planning is defined by a tuple of three elements (A,
I, G). A describes a set of actions, I represents a
initial state, and G is a goal to be achieved. Let P be
the set of all propositions that represents facts in the
world. The current state, or world, is assigned to w
and represents the subset of satisfied propositions in
P so that w ⊆ P in the world. In STRIPS, an action is
ICE-B 2006 - INTERNATIONAL CONFERENCE ON E-BUSINESS
228