(186t) Data-Driven Multi-Period Planning Model and Global Optimization for Entire Petroleum and Petrochemical Operations

Over the past few decades, the petroleum and petrochemical industry has made significant contribution to the economic development and life quality improvement. The industry converts and upgrades crude oil into a range of useful products including liquefied petroleum gas (LPG), kerosene, naphtha, gasoline, jet fuel, fuel oil and many precursors for commodity chemicals such as benzene, toluene, and xylenes (mainly p-xylenes) via different reactions and separations. However, the petroleum and petrochemical industry is facing great challenges these days from global competitions due to the declined crude quality, fluctuation of crude prices and product demands, as well as stricter environmental regulations. The difficulty in making effective planning decisions has driven the petroleum and petrochemical industry to make use of cutting-edge techniques, especially advanced mixed-integer mathematical programming techniques.

Integrated petroleum and petrochemical operations include both refinery and chemical production operations, such as crude oil blending and processing, production unit operations, and product blending and distribution [1-2]. An integrated petroleum and petrochemical operational planning involve simultaneous optimization of crude blending, selection of units and their production modes, flow connections between production units, and pooling and blending operations to satisfy quality requirements of production units, intermediates, and final products. Mathematical modelling of production units, pooling and blending operations often introduces bilinear, trilinear terms, exponential and higher order terms, which result in a non-convex mixed integer nonlinear programming problem (MINLP).

The refinery planning problem has received considerable attention since the introduction of linear programming in the 1950s. The research focused on developing different models and algorithms to solve large-scale industrial problems, leading to the development of commercial software such as RPMS (Refinery and Petrochemical Modeling System) [3], PIMS (Process Industry Modeling System) [4], and GRTMPS (Haverly Systems) [5]. While the commercially available software can be extended to model and optimize integrated petroleum and petrochemical plants, inaccuracy caused by non-rigorous linear models for product yield and property prediction in processing units may reduce the overall profitability or sacrifice product quality. Additionally, ε-global optimality cannot be guaranteed. On the other hand, nonlinear models have been proposed for product yield and property prediction in processing units during refinery planning without ensuring ε- global optimality [6-9]. A comprehensive review of refinery planning can be found in [1-2].

All the above efforts were made for solely the refinery planning without integrating petrochemical operations. However, modern refineries are often integrated with petrochemical operations to improve the site efficiency, and consequently the profit margin. Recently, Li et al [10] developed a data-driven single-period planning model for integrated petroleum and petrochemical operations. They covered in detailed how the data was initially collected and processed to build the basis for the model. The postulation of yield and outlet properties was done thoroughly by using a swing-cut approach to model crude distillation unit (CDU) and linear, quadratic, bilinear or polynomial models, or the combination of them to model the secondary unit operations such as hydrotreating units, hydrocracking units, and fluidized catalytic cracking unit. This comprehensive model has also accounted for parallel unit selection and operation mode selection. However, the operation mode selection was done using a lumping method, which may lead to the incorrect prediction of product properties, affecting the yield and property prediction for the downstream units.

In this work, a multi-period MINLP planning model is developed using the data-driven prediction correlations of product yields and properties in processing units from Li et al. [10]. The planning horizon is divided into multiple periods based on profiles of production modes and product demands. In each period, only one production mode is allowed for units with multiple operating modes such as the CDU and hydrocracking units. The objective is to maximize the total profit including revenues from product sale minus raw material and operating cost. The resulting multi-period planning model is a large-scale nonconvex mixed-integer nonlinear programming model. To solve the model to ε-global optimality, two-stage solution approach is proposed where a feasibility problem is first constructed by introducing slack variables to the yield and property prediction correlations in the original MINLP model and changing the objective to minimization of these slack variables. This feasibility model is solved using ANTIGONE/GAMS [11], whose solution indicates whether the original problem is feasible or not. If all slack variables are very close to or exactly zero, the original problem is feasible. Then, all slack variables are fixed to their values obtained from the feasibility problem, another MINLP optimisation problem having the same objective function as the original MINLP model and the same constraints as the feasibility problem is constructed and solved to ε-global optimality using ANTIGONE/GAMS[11]. Two case studies from a real petroleum and petrochemical site in China are used to illustrate the capability of the proposed multi-period planning model and postulated solution approach. Comparisons were also made between the results generated from the proposed multi-period planning model and that from the single-period planning model. The results demonstrate that the proposed multi-period planning model could generate much higher profit than that of single-period planning model, with more accurate yield and property predictions for those processing units with different operating modes.

References

[1] Shah, N. K.; Li, Zu; Ierapetritou, M. G. Petroleum Refining Operations: Key Issues, Advances, and Opportunities, Industrial and Engineering Chemistry Research, 2011, 50, 1161-1170.

[2] Khor, Cheng Seong, Varvarezos, Dimitrios, Petroleum Refinery Optimization, Optim Eng, DOI 10.1007/s11081-016-9338-x.

[3] RPMS (Refinery and Petrochemical Modeling Systems): A System Description; Bonner and Moore: Houston, TX, 1979.

[4] Aspen PIMS System Reference (v7.2), Aspen Technology Inc.; Burlington, MA, 2010.

[5] GRTMPS (Haverly Systems), http://www.haverly.com/main-products/13-products/9-grtmps.

[6] Pinto, J. M.; Moro, L. F. L. A Planning Model for Petroleum Refineries. Braz. J. Chem. Eng. 2000, 17, 575-585.

[7] Li, W. K.; Hui, C. W.; Li, A. X. Integrating CDU, FCC, and Product Blending Models into Refinery Planning. Computers and Chemical Engineering 2005, 29, 2010-2028.

[8] Alhajri, I.; Elkamel, A.; Albahri, T.; Douglas, P. L. A Nonlinear Programming Model for Refinery Planning and Optimization with Rigorous Process Models and Product Quality Specifications. International Journal of Oil, Gas, and Coal Technology 2008, 1, 283-307.

[9] Alattas, A. M.; Grossmann, I. E.; Palou-Rivera, I. Integration of Nonlinear Crude Distillation Unit Models in Refinery Planning Optimization. Industrial and Engineering Chemistry Research 2011, 50, 6860-6870.

[10] Li, J.; Boukouvala, F., Floudas, C. A., Zhao, B., Du, G., Su, X., Liu, H., Data-Driven Mathematical Modelling and Global Optimization Framework for Entire Petrochemical Planning Operations, AIChE Journal, 62, 3020-3040.

[11] Misener, R.; Floudas, C. A. ANTIGONE: Algorithms for continuous/Integer Global Optimization of Nonlinear Equations. Journal of Global Optimization, Journal of Global Optimization, 2014, 59, 503-526.