Resolving conflicting objectives in factory scheduling

A primary source of difficulty in constructing good job-shop schedules stems from the conflicting nature of the domain´s objectives. Traditional approaches to scheduling have tended to reduce the complexity of the problem by considering only a small subset of objectives. However, the author demonstrates that this approach is unsatisfactory and that it is necessary to reason dynamically about global satisfaction to produce good, balanced schedules in realistic environments. The paper describes the current state of research into, and implementation of, knowledge-based scheduling techniques. It also details an investigation into the role of predictive techniques in building and maintaining `good´, balanced schedules. Several different ways of constructing, propagating and using predictive knowledge, such as a simple capacity plan or a more complicated probabilistic analysis are outlined and test results compared. Finally, an approach to resolving the problem of conflicting objectives is presented. The approach combines the information provided by a probabilistic analysis with the ability to represent the influencing constraints and reasons about the dependencies between them