Integrated scheduling optimization in the crude oil refinery industry: from crude oil unloading to fuel deliveries.

• Autor(a) principal: Franzoi Junior, Robert Eduard
• Data de Publicação: 2021
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/3/3137/tde-01062021-101757/
• topic: Crude oil refining; Heuristic approaches; Heurística; Modelagem e otimização; Modeling and optimization; Modelos surrogados; Online scheduling; Petróleo (Refino); Programação de produção online; Surrogate modeling

The crude oil refinery scheduling optimization is a complex and challenging problem because of its large-scale and complex-scope non-convex MINLP formulation. Three main concepts have been adopted in both industry and academia to handle this issue. First, a simplified formulation is typically considered, which does not include all the processing units, tanks, flows, and variables from the real industrial problem. Second, the refinery scheduling formulation is broken down into subproblems to be hierarchically solved. Third, simulation-based instead of optimization-based approaches are still employed due to the intractability of such formulation. However, the recent advancements in decision-making modeling, computer-aided resources, and solution algorithms allow the modeling and optimization of previously intractable problems, provide resources for novel real-time industrial applications, and open opportunities for the development of novel and improved modeling and optimization strategies. The research topics addressed herein focus on handling complex formulations typically found in crude oil refinery scheduling applications. The novelty of this research consists of modeling and optimizing a complete crude oil refinery scheduling problem, including decomposition approaches for handling intractable formulations, improved network designs for blending and processing operations, rescheduling strategies for online applications, and surrogate modeling for integrated optimization environments. Decomposition approaches are useful for building simpler and tractable formulations from complex and large-scale problems. Improved processing and blending designs provide more accurate predictions, production flexibility, and increased economic value for the process. Modeling and solving heuristics are used to significantly reduce the computational effort by limiting the optimization search space in constructive rolling horizon strategies and by introducing iterative relaxations on mixed-integer linear programming problems. Rescheduling and parameter updating strategies mitigate plant-model mismatches by effectively handling uncertainties and disturbances, reducing inaccuracies, maintaining the state of the system updated, and providing a systematic fashion for online applications. Surrogate models can effectively replace complex formulations in order to allow the integration of unit-operation models within refinery scheduling optimization environments. The formulation and methodologies addressed herein are coherent with large-scale and complex-scope industrial applications in terms of applicability, operational constraints, refinery economics, and problem complexity and size. The results indicate that complex non-convex MINLP refinery scheduling formulations can be efficiently solved by utilizing decomposition, heuristic, machine learning, and rescheduling strategies,which would potentially provide improved modeling and optimization capabilities for real industrial applications.