(A) knowledge-based linear programming model formulation system

Abstract

Though there have been numerous research on the algorithmic efficiency of linear programming (LP) models, there were not enough attention about the research on the automatic formulation of LP models. Recently, there are a few studies that attempt to support the modeling stage using the Artificial Intelligence techniques. For this approach, the knowledge engineers adn LP modeling experts should, of course, prepare knowledge base beforehand. Although the preparation is not free, the prepared knowledge base can be shared by many novice users whose models can be built as a subset of the knowledge base. We propose a method that the LP models can be formulated via the semantic level dialogue of context identification. For example, for the formulation of the models about production management, the modeler merely tells what to produce, what the facilities are, etc; then the relevant LP model will be generated. To support such a namely automatic formulation, the knowledge are prepared in frames and constraints. The frames are good to represent the domain context. And the constraints represent the equations adn inequalities. Constraints can be represented as a network which is semantically meaningful. The prototype UNIK-OPT (UNIfied Knowledge OPTimization) has developed to realize the proposed approach. The illustrated domain of application is the production management of petroleum industry. UNIK-OPT has three levels of views about the LP models: semantic, notational, and tabular views. The UNIK-OPT is implemented using LISP on IBM PC. The framed-based data base is attached to the UNIK-OPT for the maintenance of data consistency.