(55b) Hinging Hyperplanes Crude Oil Mixing Model for Production Planning Optimization

Refinery feed, usually a mixture of several crude oils, is separated via crude distillation unit (CDU). Crude true boiling point (TBP) distillation curve determines the amount of products which can be obtained via CDU separation. TBP curve of the mixed crude feed is calculated by blending pseudocomponents from each individual crude; the blending calculation introduces a lot of nonlinear terms and makes it difficult to optimize the crude selection and the downstream refinery operation. This work proposes a continuous piecewise linear (CPWL) model to approximate the crude blending, thereby eliminating the nonlinear blending terms. A hinging hyperplanes (HH) model is employed to formulate the CPWL model. Additionally, its parameters are computed by a two-layer perceptron, which is built to simulate the HH model. This leads to a mixed-integer linear model suitable for crude oil selection and cut point optimization. The proposed model enables accurate computation of the mixed crude feed TBP curve, while eliminating nonlinear blending terms. Accuracy of the proposed approximation model has been tested on mixtures of two crudes and five crudes. It is shown that a high approximation accuracy is attained by having a sufficient number of hinges which correspond to the nodes. At the same time, the model still retains an easy to solve (small size) MILP problem in the feedstock selection optimization. Since pseudocomponent based mixing model leads to an MINLP formulation, its execution times are significantly higher than the hinging hyperplanes based mixing model. At the same time, the high accuracy of the hinging hyperplanes model ensures that its use in the planning models will lead to the same results as the pseudocomponent based crude mixing model. Hence, it is suitable for inclusion in refinery planning models dealing with optimal crude selection.