Although there have been numerous mathematical applications to the problems of petroleum industry, there have been rare ones to the production scheduling for oil refinery. Most of the mathematical models have failed in this problem because of the inherent nonlinearity among the relevant decision variables and the need of multiperiod optimization. Therefore, it is natural to try to solve this problem by a heuristic method. Under this rationale, an architecture of a knowledge-based production scheduling system for oil refinery coupled with the optimization models, IORS (Intelligent Oil Refinery Scheduling system), is proposed. A characteristic of IORS is the fact that the scheduling problem is decomposed into knowledge-based part and optimization part.
The IORS adopts two types of knowledge representation scheme: frame for declarative knowledge and production rule for expert``s scheduling heuristics. The rules are classified into several groups to improve the efficiency of pattern matching. The prototype of IORS is developed using UNIK-FRAME and LISP language on a IBM PC, and is applied to an existing refinery plant. According to the limited experiment, IORS generates feasible schedule effectively.