Optimization of monthly crude oil scheduling in refineries with ship arrivals

Abstract

The thesis focuses on optimizing crude oil operations scheduling in a refinery with maritime access, addressing the problem through both deterministic and stochastic approaches, and considering the gaps identified in the literature for each. For the deterministic approach, the refinery under study is supplied with different types of crude oil via vessels, and it must carry out several operations to meet the specified demand. These operations involve coordinating ship arrivals, managing crude oil inventory, and determining the feeding schedule for crude distillation units (CDUs) and downstream processing units. A key feature of this problem is that tanks can store crude blends, introducing nonlinear and non-convex constraints. Additionally, tanks are classified based on the type of blend stored, and the scheduling horizon is one month, resulting in a large-scale model that is difficult to solve within operationally acceptable times. To address this challenge, a mathematical programming model is developed using a continuous-time formulation with global time points, which represent decision points for evaluating limited resources. The model supports decisions such as when to unload vessels, which tanks to use, how to feed the CDUs, for how long, and in what quantities. Given the model size and nonlinear nature, a novel solution strategy is proposed. It combines a piecewise linear approximation technique using planes to linearize the product of non-negative continuous variables, with a temporal decomposition scheme that iteratively solves an aggregate and a detailed model. A case study is solved with real refinery data, and the results are analyzed by means of graphs (Gantt diagrams, evolution of mixture properties) and tables with model statistics and computation times. The results indicate that this methodology allows obtaining high-quality solutions in times compatible with the operational requirements. The thesis also addresses the problem from a stochastic perspective, motivated by the uncertainty often present in operations scheduling, such as variability in ship arrival times, which can compromise the feasibility of a deterministic plan. A two-stage stochastic programming model is developed to capture this uncertainty. Although the model is aimed at short-term scheduling and simplifies some aspects compared to the deterministic model, it maintains sufficient complexity to serve as a proof of concept for evaluating the impact of uncertainty. The stochastic model is further extended to include risk management by incorporating the Conditional Value-at-Risk (CVaR) as the objective function. Performance is assessed using metrics such as the Expected Value of Perfect Information (EVPI) and the Value of the Stochastic Solution (VSS), in order to compare the potential advantages of the stochastic approach over the deterministic one, and to analyze solutions under different levels of risk aversion. The results show that the two-stage stochastic programming model yields more robust schedules by allowing current decisions to adapt to future conditions. Moreover, including CVaR in the objective function helps mitigate the effect of extreme scenarios, thus enhancing risk management in crude oil scheduling.