(761h) Robust Refinery Planning Under Exogenous and Endogenous Uncertainty

This paper proposes a modeling and optimization framework for supporting the decision-making of optimal feed acquisition in refinery planning while simultaneously considering both exogenous and endogenous sources of uncertainty. The presented approach consists of two distinct decision stages that are separated in terms of time-scales: (a) a strategic stage in which we determine the optimal feedstock slate (such as crudes or chemicals feeds) to purchase for the long-term horizon (using a contract approach - stage 1), and (b) a tactical stage in which we determine the optimal feedstocks to purchase on a short-term horizon basis (utilizing spot market purchase opportunities - stage 2). These stages are evaluated at separate time intervals throughout the planning horizon. Unlike existing approaches such as stochastic programming or multi-stage recourse, that often require complex decomposition strategies, the proposed algorithm allows hedging against future risk while maintaining computationally-tractable problem instances using the principles of Monte-Carlo simulation and discreet event simulation. Specifically, the uncertainty considered in this work can be (a) exogenous, such as supply, demand and pricing, as well as (b) endogenous, such as asset availability that is directly calculated from reliability and maintenance considerations, that are typically the outcome of discrete-event simulation. In principle, the proposed two-stage approach can be used with both exogenous and/or endogenous uncertainties for both short and long term decision making horizons. In this work, we utilize the use of chance-constrained optimization for the determination of robust crudes (price-resilient) for contract purchase in stage 1, and perform a breakeven analysis under equipment reliability and availability uncertainty in stage 2. We apply this strategy to an industrial case study involving a refinery model in order to assess tangible economic benefits and evaluate the effectiveness of the proposed algorithm. The numerical results presented here indicate that this systematic methodology provides significant economic benefits and it far outperforms the deterministic approach, leading to an increase in the expected refinery profit of more than $20 million, on average, per year.