Can Multiple Criteria Methods Help Production Scheduling?

Abstract

Operational Research has contributed in a number of ways to production scheduling; the development of algorithms to identify optimal solutions to simple problems; the development of heuristics to quickly find good solutions to complex problems; and the use of visual interactive modelling to provide a flexible and easy to use tool for production schedulers.

These contributions have had significant practical impact. However, some difficulties remain and one of these is related to the multiple criteria nature of the problem. Commonly used heuristics tend to be based on a single criterion rule, a very simple example being “Schedule the job with shortest processing time first”. The heuristic may produce a schedule which performs well against one objective, for example, to minimise the number of jobs late, but does not meet the scheduler’s requirements against other criteria, for example, minimising stock, or controlling maximum lateness. However, although the scheduler can easily identify shortcomings of a schedule with the support of appropriate software, he or she may be unable to identify what changes to make to improve the situation.

The aim of our research is, through the development of a multiple criteria scheduling heuristic, to provide the scheduler with a control mechanism. This will allow him or her to indicate how they would like the solution to change and by appropriately adjusting the heuristic produce a new schedule which moves towards the specified objectives.

As a first step we have been working with a simple multiple criteria heuristic, implemented as a visual interactive decision support system, to investigate the feasibility of such a control mechanism. In this paper we will demonstrate this system and report on the results of some initial experimental work. The next stage of the research will depend on the results of this experimental work; if the multiple criteria heuristic used yields the desired results we can move towards implementing the control mechanism, for which we will explore the use of expert systems and neural networks; if the simple heuristic proves to be ineffective we will have to investigate a more complex approach. We will report on progress at the Conference.