Customer oriented production planning at an integrated steel works

Abstract

Developers of decision support systems for production planning domains often encounter ill-structured problems that arc characterised by uncertairrty. This report describes a knowledge-based decision modelling approach which successfully addressed these issues at a steelmaking plant. All planning knowledge was reduced to a set of hard and soft constraints explicitly derived from the relevant domain experts. A prototype system was designc.l which allowed the operators to vary the relative priorities of the soft construints according to the often unstable requirements of the various stukeltolders in the planning decision process, Soft constraint inferencing was modelled in two ways - a binary heuristic approach and a fuzzy constraint analysis. The tWI) approaches arc evaluated and compared. The binary heuristic approach enforced a rigid semantic partitioning of the constraint spaces. The cOllsequelll/}.I!{IIIl'lltatioll of the problem resulted in a prohibitively complex system, which provided neither consistent nor reliable decision support. The fuzzy constraint analysis provided significantly improved results. It allowed an intuitive representation of soft constraint semantics. Uncertainty was represented as all intrinsic part of the decision model. The rule base was significantly reduced and the inference rules were closet' to the manner in which the domain experts reason about the problem. The operators were able to effectivel» impose their variable decision criteria on the model.