Title: Synthesis of Schedulers for Planned Shutdowns of Power Plants


Authors: Carla P. Gomes (gomes@ai.rl.af.mil), Douglas R. Smith
(smith@kestrel.edu), and Stephen J.  Westfold (westfold@kestrel.edu)


Presenter: Carla P. Gomes


We describe the synthesis of efficient schedulers for planned
shutdowns of power plants for refueling and maintenance (outages),
using an automated programming tool, KIDS.  Currently, the utility
industry has no automated tools to generate schedules that are both
safe and resource-efficient.  We focused on safety constraints since
they are critical in this application.


There are several aspects of this project that go beyond previous
applications of KIDS to scheduling problems.  First, scheduling of
outages of power plants has a planning-like character
since the scheduler needs to represent and maintain the complex state
of the plant at all times considered during the scheduling process.
Second, the particular safety constraints we considered required
scheduling a pool of resources in the presence of time windows on each
activity.  To our knowledge the control and data structures that we
developed for handling such a pool are novel.


In terms of design knowledge, the outage scheduling problem is modeled
as a {\it constraint satisfaction problem} and the synthesized
algorithm is an instance of {\it global search} with {\it constraint
propagation}. The derivation of specialized representations for the
constraints to perform efficient propagation is a key aspect of our
approach.  In addition, {\it finite differencing} complements
constraint propagation by efficiently maintaining the state of the
world.