Simulation and optimization methods as decision support tools for operation of oil refinery hydrogen networks
- Galán Prado, Aníbal Santiago
- César de Prada Moraga Directeur
- Gloria Gutiérrez Rodríguez Co-directrice
Université de défendre: Universidad de Valladolid
Fecha de defensa: 20 janvier 2021
- Carlos Bordóns Alba President
- Rogelio Mazaeda Secrétaire
- Lars Imsland Rapporteur
Type: Thèses
Résumé
The process industry has been at the forefront of manufacturing technology advances, systematically aiming for sustainable process improvements. Generally, improvements come as a result of decision-making processes which may be backed on information from various sources. In fact, decision making in manufacturing operations is indeed complex, as it accounts for different scopes and timescales depending on the aimed activities ranging from strategic business management decisions (long term decisions) to basic regulatory control of the process (automatic actions). Despite that, actual deployment of integrated decision support systems remains atypical in process industries, which constraints the potential of optimal operation of assets. This thesis focuses primarily on advanced and supervisory control tools, which typically translates into plant-wide decisions at timescales from minutes (advanced control) to hours (supervisory control). In this context, this thesis centers the discussion on crude oil refinery hydrogen networks operations, especially considering change of condition mitigation measures to support operators' decision-making process. Hydrogen in oil refineries is used for sulfur removal from intermediate products to produce commercial fuels (e.g.: diesel, gasoline) and increase crude oil yields. Therefore, refinery hydrogen networks operation demands maximization of process units loads (benefits) minimizing hydrogen production (costs), subjected to uncertainties of the equipment and bounded to operational and safety constraints. In particular, this thesis uses Petronor refinery in Bilbao, Spain, as case study. The main aim of this thesis is to study and develop a decision support tool for refinery hydrogen network operators. For this purpose, it is proposed an integrated decision support framework for hydrogen network operators under uncertain conditions, which combines process information, model predictive control, optimization and simulation tools. Firstly, a first principles hydrogen network dynamic simulation library is described and discussed towards its utilization as the backbone of decision support tools with focus on the architecture. Furthermore, an example of architecture is proposed and analyzed towards the use of this library for developing enhanced simulation environments. The discussion ends with thoughts, based on an example, about leveraging dynamic simulation into real-time operation environments. Secondly, a simulation-based decision support tool called real time reconciled simulation is introduced. The study describes real-time reconciled simulation (RTRS), and analyzes its usefulness as decision-making tool for process operators, especially under unexpected process changes. The proposed methodology and architecture is implemented in two case studies in the context of an oil refinery hydrogen network, both plant and network levels are considered. A what-if analysis is conducted on case studies, assessing two feasible mitigation actions for each case baseline condition. The focus of the discussion is, nevertheless, on the methodology itself and its general features as decision support tool. It is highlighted the fact that RTRS complements in a straightforward manner other control operation tools such as model predictive controllers (MPC) and real-time optimizers (RTO). Therefore, it may add to any decision support framework an open-loop component with parameter estimation and forecasting capabilities. Moreover, its potential for training and integration within other tools packages is discussed. Thirdly, the problems associated with the implementation of a real-time optimization (RTO) decision support tool, for the operation of a large scale hydrogen network of an oil refinery is addressed. In addition, a formulation which takes into account the stochastic uncertainty of hydrogen demand, due to hydrocarbons quality change, is described and further studied, focusing on its utility in the decision-making process of operators. An integrated robust data reconciliation, and economic optimization, considering plant-wide uncertain parameters is presented and discussed. Moreover, stochastic uncertainty in hydrogen demand is assessed for its inclusion within the RTO framework. A novel approach of the decisions stages at hydrogen producers and consumers is proposed, which supports the formulation of the problem as a two-stage stochastic non-linear program. Representative results are presented and discussed, aimed at assessing the potential impact in the hydrogen management policies. For this purpose, the value of the stochastic solution, perfect information, and expectation of the expected value are analyzed. Complementarily, a risk-averse formulation is presented (value-at-risk and conditional-value-at-risk) and its results compared against the risk-neutral formulation. Fourthly, the integration of multiple decision support tools is presented and discussed in the context of what is named decision support frameworks (DSFs). These are the natural environment of decision-making tools, however in practice these tools reside in different silos across multiple systems within company. This configuration challenges information and data exchange amongst tools in a transparent manner, which can result in inconsistent solutions due to different applications using different data sets for the same purpose. In order to address these issues and provide enhanced decision-making support across operations, a DSF architecture is proposed and studied. The DSF architecture features previously presented tools such as RTO and RTRS, while it introduces the data management system role, the digital twin role, online and offline simulation and other features. The discussion is focused on how the architecture would improve decision makers' capacity of making complex decisions supported by updated information from across the business, with especial interest in process operations. Furthermore, DSFs promote enhanced process knowledge and skills transfer, due to their ease-of-access to consistent information along with their forecasting and assessment capability over multiple operation alternatives (e.g. What-if analysis). In addition, the proposed DSF architecture considers tailored models, which are supported by the library of models under the scope of the digital twin. The final section of this thesis consists of a summary of the conclusions of each section, along with future challenges and open issues going forward. This section ends with a list of publications and contributions of this thesis.