(343f) An Integrated Data-Driven Modeling & Global Optimization Approach for Production Planning Under Uncertainty

The tight competition, environmental regulations, and lower profit margins are some of the factors that drive the petrochemical industry to make planning and scheduling operations more efficient over the last few decades. Planning and scheduling have been topics of special interest for many process systems engineers as constrained optimization problems and the refinery planning problem has received considerable attention since the introduction of linear programming in 1950s [1]. Complex production facilities such as petroleum refineries are highly interconnected and the objectives of most of the individual units are conflicting and optimizing a single unit’s operation can end up the overall operation to be suboptimal if not infeasible [2], thus making a holistic approach both necessary and advantageous.

Traditional production planning approaches rely on LP principles with fixed-yield planning models, even though refinery operations such as distillation, processing, and pooling are highly nonlinear in nature. While there are commercially available highly detailed mathematical models for simulation of processing units, such detailed models cannot be used in enterprise-wide optimization-based approaches due to high computational expense. Recent studies showed that, data-driven modeling offers a promising way to obtain inexpensive nonlinear models, which can relate relevant inputs to relevant outputs to describe each individual process accurately, and therefore making a plant- or enterprise-wide optimization-based approach realizable [3,4]. The increase in computational power of global optimization solvers such as ANTIGONE and BARON can open the door to planning frameworks that are formulated as NLPs or MINLPs [5,6].

In this work, we are proposing a framework for integrated data-driven modeling and global optimization to solve production planning problems. In the first step of the work, we organize, analyze, and process the real plant data provided by Hyundai Oilbank’s Daesan Refinery located in South Korea. Linear and nonlinear models of different complexities are trained, validated, and then compared to find the best models to describe the processes. In the second step of the work, we create a superstructure containing all possible connections and operating modes in the refinery. The resulting multi-period planning model is a non-convex nonlinear optimization model (NLP), which is solved to e-global optimality using ANTIGONE. Results of several case studies are provided to illustrate the efficiency of our proposed model and global optimization approach compared to the actual plant operation. Multi-period formulation is compared with single-period formulation, and the uncertainty in demands and prices are addressed by robust optimization.

1. CA Floudas, AM Niziolek, O Onel, LR Matthews, 2016, Multi-Scale Systems Engineering for Energy and the Environment: Challenges and Opportunities, AIChE Journal, 62 (3), pp. 602-623.
2. JM Pinto, M Joly, LFL Moro, 2000, Planning and scheduling models for refinery operations, Computers and Chemical Engineering, 24 (9-10), pp. 2259-2276.
3. F Boukouvala, CA Floudas, 2017, Argonaut: Algorithms for global optimization of constrained grey-box computational problems, Optimization Letters, 11(5), pp. 895-913.
4. J Lie, X Xiao, F Boukouvala, B Zhao, G Du, X Su, H Liu, CA Floudas, 2016, Data-Driven Mathematical Modeling and Global Optimization Framework for Entire Petrochemical Planning Operations, AIChE Journal, 62 (9), pp. 3020-3040.
5. R Misener, CA Floudas, 2014, ANTIGONE: Algorithms for coNTinuous Integer Global Optimization of Nonlinear Equations, Journal of Global Optimization, 59 (2–3), pp. 503–526.
6. M Tawarmalani, N Sahinidis, 2005, A polyhedral branch-and-cut approach to global
   optimization, Mathematical Programming, 103(2), pp. 225-249.