(472a) An Empirical Model for Crude Distillation Units in Petroleum Refinery Planning

Petroleum refinery planning models require accuracy in the modeling of the processes involved, since the results from planning models should be achievable in the refinery operation. Inaccuracy caused by non-rigorous models may lead to suboptimal feedstock selections and unrealistic operating plans. Nevertheless, the accuracy of product yield and quality predictions in petroleum refinery planning models is a difficult task because of the extremely complex economics of petroleum refinery (crude oil can be processed in many available refinery configurations using various technology options in accordance with market demand and specifications of blended products) and the inherent nonlinearity of the processes.

The crude distillation unit (CDU) is one of the most important process units in a refinery. The CDU fractionates the crude oil into different products such as light end hydrocarbons, gasoline, kerosene, diesel and atmospheric residue. Some of these products can be sent to the blending system directly or can be processed in downstream units to be converted into final products. Thus it is essential to ensure the accuracy of the CDU model in refinery planning models. Lately, the approaches to model the CDU reported in the literature include the swing cut model (Zhang, 2001) and the weight transfer ratio (WTR) model (Li et al, 2005). The swing cut model is commonly used in some commercial software for refinery planning such as PIMS® (Aspentech) and RPMS® (Honeywell). In this model, the properties of CDU fractions are assumed to be constant across their temperature ranges; this is not true since the size of the swing cut can vary from 5 to 7 vol % of the overall crude fed and property distributions in CDU fractions are highly nonlinear. On the other hand, the WTR model is not used in commercial software, but can be very efficient when the true boiling point (TBP) curve of the crude oil fed into the CDU is known. However, most refineries may be processing more than one crude oil and the TBP curve of the crude oil (mixed crude oil) fed into the CDU is unknown and thus the WTR model can not be implemented.

In this study, a nonlinear empirical model for CDUs was developed. Differently from previous works, the decision variables in this model are the TBP cut points between each two adjacent fractions which are decision variables easier to be directly implemented in the refinery operation and the flow rate of each crude oil that is fed into the CDU, therefore, it is not necessary to know the TBP curve of the mixed crude oil fed into the CDU. The lower and upper bounds for the TBP cut points were estimated by using the procedure suggested by Watkins (1979); the required data to apply this procedure are the ASTM D86 100 % specification for each fraction and the separation grade gap (5-95) ASTM between adjacent fractions. In order to reflect the influence of TBP cut points on the product qualities, the properties of CDU fractions were correlated to their mid-point volume transfer ratios. The yield predictions from this model were compared with rigorous simulations carried out in HYSYS® (Aspentech). The HYSYS model for the CDU consists basically of a main column with 29 theoretical trays, 3 pump-arounds and 3 side strippers. Two crude oils (light crude and heavy crude) were fed into the CDU, the ASTM D86 100 % specifications and the separation grades gap (5-95) ASTM being used as specifications in the HYSYS model. Three cases were compared and the absolute errors in the yield prediction were found to be less than 4.2 vol %, furthermore, when bias factors were added to the empirical model predictions, the absolute errors in the yield predictions were reduced to less than 0.52 vol %. These bias factors reflect the effect of operational and design variables that were not considered in the empirical model, since the HYSYS model has more degrees of freedom than the empirical model.

Finally, the CDU model was implemented in a petroleum refinery planning model detailed in the literature (Li et al, 2005). The refinery planning model includes a CDU, a fluid catalytic cracking (FCC), a gasoline blending (GB) and a diesel oil blending (DB) units. Two crude oils are available to feed the CDU where gross overhead (GO), heavy naphtha (HN), light distillate (LD), heavy distillate (HD) and bottom residue (BR) fractions are produced. The major refinery products are 93 # gasoline, 90 # gasoline, -10 # diesel oil, 0 # diesel oil and heavy oil. MTBE is available as a feed to GB in order to satisfy the quality requirements of 93 # gasoline and 90 # gasoline. Crude oil prices were estimated by using the quality discount technique (Bacon & Tordo, 2004) for API and sulfur content (wt %), 16 crude oils (from both OPEC and non-OPEC countries) were used to estimate the average quality discount for each quality.

The planning model was implemented in the GAMS (GAMS 22.9) modeling language; three NLP solvers (CONOPT 3, IPOPT and MINOS) available in the GAMS platform were tested (using default settings) with 100 random starting points. It was found that the solvers CONOPT 3 and IPOPT were very efficient at solving the problem (these solvers reported local solutions for all starting points) while MINOS did not report a local solution for 23 of the 100 starting points. All local solutions were found to be the same.

Keywords: nonlinear, empirical model, CDU, refinery planning.

References

[1] Bacon, R., & Tordo, S. (2004). Crude Oil Prices: Predicting Price Differentials Based on Quality. Private Sector Development Vice Presidency. Washington: The World Bank.

[2] Li, W., Hui, C.-W., & Li, A. (2005). Integrating CDU, FCC and product blending models into refinery planning. Computers and Chemical Engineering , 29, 2010?2028.

[3] Watkins, R. (1979). Petroleum Refinery Distillation (2nd ed.). Houston,TX: Gulf Publishing Co.

[4] Zhang, J. (2001). A Level-by-Level Debottlenecking Approach in Refinery Operation. Industrial and Engineering Chemical Research , 40, 1528?1540.