Modeling of crude oil blending via discrete-time neural networks

Abstract

Crude oil blending is an attractive solution for those refiners who have the ability to blend different crude types to provide a consistent and optimal feedstock to refinery operations. Optimal crude purchasing is an effective method to improve refinery profits. In general the blending rule is nonlinear, it can be regarded as a linear mixing rule adding a nonlinear term. Crude oil blending is an optimization operation based upon real-time analyzers and process knowledge (6). A mathematical model for crude oil blending is needed to address uncertainties in blending operation, real-time optimization (RTO) has been proposed (20). The main drawback of RTO is that it cannot provide optimal set-points from large amounts of history data. © 2010 Springer-Verlag Berlin Heidelberg.