EXCALIBUR
Adaptive Constraint-Based Agents in Artificial Environments

[JOB-SHOP]   [Realization]   [Results]   [Constraint Weights]

[ Please note: The project has been discontinued as of May 31, 2005 and is superseded by the projects of the ii Labs. There won't be further updates to these pages. ]

Results

(Related publications: [PUBLink] [PUBLink])

The figure below shows experiments with different horizons. The global search control selects a constraint with a probability proportional to the constraint's costs. The schedule always contains 50 jobs ( 50 TCs), each of them with five tasks, and there are five machines ( 5 ARCs). Every tenth of a second there is a job removal/addition.

According to a computation using refinement/global search by the ConPlan system [PUBLink], the minimal horizon for a maximal consistent solution varies around 6,000, depending on the currently active jobs. With a horizon of 2,000, the topology of the search space is so flat that any improvement effect is close to pure noise. A more complete picture is given in the figure below. One point represents the inconsistency averaged over 10 seconds of runtime.

In order to study the search behavior until complete satisfaction was achieved, no job removal/addition was done for the rest of the experiments. There are 10 jobs ( 10 TCs) with 10 tasks, 10 machines ( 10 ARCs) and a horizon of 2,300. The start inconsistency is about 50,000.


[JOB-SHOP]   [Realization]   [Results]   [Constraint Weights]

For questions, comments or suggestions, please contact us.

Last update:
May 19, 2001 by Alexander Nareyek