(745b) Robust Refinery Planning Under Uncertainty

Refinery planning is the problem of determining what crudes to purchase and how they will be processed in order to meet all demand requirements and specifications while optimizing operations.  Decisions on which crudes to run are typically made in two stages, contract and spot, with contract decisions having to be made in advance of operations and before uncertainty in spot crude or product prices can be resolved.  Crude purchased on contract tends to represent the majority of crude processed at a refinery.  This forms a basic hedging strategy that ensures: (i) feed availability, (ii) cost assurance, and (iii) potential price discount.  This of course limits the potential profit upside and it also depends on the location.  Coastal refineries with more optionality tend to have a larger portion of their feedstock purchased from the spot market.

Refinery planning is modeled as a mixed integer, non-linear optimization problem that can contain tens to hundreds of thousands of equations and variables.  Due to the size of the models, making optimal contract purchase decisions using stochastic programming techniques can quickly become computationally infeasible.  In this paper, we explore ways to make the two stage crude purchase decision process under uncertainty computationally attractive.  In particular, we propose a specialized partition strategy that categorizes, samples and optimizes over multiple uncertain parameters such as benchmark crudes, crack spreads, and key equipment availability, while maintaining a computationally tractable size problem.  The algorithm occurs in two stages, deciding the crudes to purchase on contract in the first stage and deciding on spot crude purchases in the second stage.  At each stage of the optimization a curve demonstrating the probability of outcomes is generated for the user to better understand the risk of a decision set and to potentially identify alternate solutions.  Finally, we apply this approach to a particular refinery plan where we demonstrate the benefits of the proposed strategy.