Crude Selection Integrated with Optimal Refinery Operation by Combining Optimal Learning and Mathematical Programming

Abstract

Crude oil procurement is an important step in refinery management as a large number of crude oil types of varying price and quality are considered. The crude quality is a dominant factor determining the quantity and quality of final products, and the overall operating costs. Thus the selection should be done carefully by considering its impact on the overall refinery operation. The main complication is that significant uncertainties exist on the crude properties before they are actually purchased and processed. Hence, the overall operating cost for each crude type is a random variable. In this study, a decision-making strategy for the crude selection and refinery operation is introduced by combining optimal learning and mathematical programming. A decision policy for crude valuation and selection is obtained by optimal learning based on the knowledge gradient algorithm with correlated beliefs. In the overall decision model, the operational variables are assumed to be determined by solving a LP problem. The uncertainty about the crude quality is propagated through the operation model, and the evaluative information on the operating cost is continuously fed back for improving the crude selection policy. The performance of the proposed approach is verified through some case studies reflecting the real refinery situation.