Reactive Scheduling Systems

Chew Ern

Abstract

Ths project develops a reactive scheduling system for a highly dynamic and unpredictable environment. An unique reactive scheduling system based on GA (genetic algorithms), nicknamed Genetic Reactor, was developed in the project. This is the first reactive scheduling system in the world based on GA. Genetic Reactor, written in Visual C++, is able to aid the decision maker in reactive scheduling even when there is little past knowledge to draw upon, a task that is of considerable difficulty for existing reactive scheduling systems.

A methodology to quantify schedule stability that takes into account the human expert's preference for change in the existing schedule was developed. This permits a bi-criteria optimization approach for reactive scheduling. The schedule stability criterion, a principal feature of any reactive scheduling system, is incorporated into the objective function, a break from the usual practice. This facilitated the exploitation of the parallel search capability of GA to generate the efficient frontier swiftly.

A GA scheduler was written to search for near-optimal schedules. To evaluate the solutions obtained using the GA scheduler,the solutions from GA were compared with the global optimal solutions. A mathematical programming model was created to generate the global optimal solutions. The GA scheduler is found to be capable of producing near-optimal solutions within tractable time.