A surrogate model approach to refinery-wide optimization
Slaback, Dale D
MetadataShow full item record
The techniques currently used to perform refinery-wide optimization can give results that are inconsistent with the overall objectives of the refinery. A full-scale nonlinear refinery-wide optimization approach can accurately predict the overall refinery optimum, but suffers from a large computational requirement. In this study, a surrogate model approach is applied to the refinery-wide optimization problem. The surrogate model approach to large-scale optimization involves building both detailed and approximate models for all of the processing units in the refinery. The detailed models developed in this study are rigorous first-principles models involving the material and energy balance equations. The surrogate model approach to large-scale optimization can employ approximate models of any form. However, selection of the proper form for the approximate models can greatly increase the efficiency of the optimization problem. In this study, fixed physical property phenomenological models are used as the approximate models. By fixing the values of the stream enthalpies and vapor-liquid equilibrium constants, the total number of equations in the process models is greatly reduced. This choice of approximate model form also guarantees that, at convergence, the results of the detailed and approximate models will be identical. The ASCEND IV modeling language is chosen for creating the detailed and approximate models in this study. This modeling platform provides significant advantages over a standard programming language such as Fortran. In addition to having a graphical user interface, the ASCEND software also contains an integrated solver and optimizer, making implementation of the optimization procedure more straightforward. By combining the models of each unit of the refinery together, a refinery-wide model is created. Using the CONOPT optimization routine in ASCEND, the refinery-wide optimization problem can be solved. The optimization results obtained in this work are consistent with the refinery-wide optimization results presented by Li (2000). For the refinery model created in this study, the surrogate model approach decreased the required solution time by nearly an order of magnitude. An optimization was also performed for a refinery in which some product was recycled back to the crude unit. By adding this recycle stream, the system of equations was made much more complex, with each unit being affected by all others. In this case, a dramatic reduction in the optimization solution time was also observed. The refinery model in this study contains 32 decision variables and 63 constraints. An industrial-scale refinery model would be much larger, perhaps including 150 decision variables. The solution time reduction using the surrogate model approach increases with the number of decision variables. Therefore, it is projected that the time reduction for an industrial-scale refinery model could be substantially larger than for the model used in this study. The speedup obtained using the surrogate model approach would decrease the solution time for refinery-wide optimization from several days to only a few hours. By decreasing the solution time, the surrogate model approach provides a method for performing refinery-wide optimization in an industrial setting.