Crude refining is one of the most energy intensive industrial operations. The large amounts of crude processed, various sources of inefficiencies and tight profit margins promote improving energy recovery. The preheat train, a large heat exchanger network, partially recovers the energy of distillation products to heat the crude, but it suffers of the deposition of material over time – fouling – deteriorating its performance. This increases the operating cost, fuel consumption, carbon emissions and may reduce the production rate of the refinery.
Fouling mitigation in the preheat train is essential for a profitable long term operation of the refinery. It aims to increase energy savings, and to reduce operating costs and carbon emissions. Current alternatives to mitigate fouling are based on heuristic approaches that oversimplify the representation of the phenomena and ignore many important interactions in the system, hence they fail to fully achieve the potential energy savings. On the other hand, predictive first principle models and mathematical programming offer a comprehensive way to mitigate fouling and optimize the performance of preheat trains overcoming previous limitations.
In this thesis, a novel modelling and optimization framework for heat exchanger networks under fouling is proposed, and it is based on fundamental principles. The models developed were validated against plant data and other benchmark models, and they can predict with confidence the main effect of operating variables on the hydraulic and thermal performance of the exchangers and those of the network.
The optimization of the preheat train, an MINLP problem, aims to minimize the operating cost by: 1) dynamic flow distribution control, 2) cleaning scheduling and 3) network retrofit. The framework developed allows considering these decisions individually or simultaneously, although it is demonstrated that an integrated approach exploits the synergies among decision levels and can reduce further the operating cost. An efficient formulation of the model disjunctions and time representation are developed for this optimization problem, as well as efficient solution strategies. To handle the combinatorial nature of the problem and the many binary decisions, a reformulation using complementarity constraints is proposed. Various realistic case studies are used to demonstrate the general applicability and benefits of the modelling and optimization framework. This is the first time that first principle predictive models are used to optimize various types of decisions simultaneously in industrial size heat exchanger networks.
The optimization framework developed is taken further to an online application in a feedback loop. A multi-loop NMPC approach is designed to optimize the flow distribution and cleaning scheduling of preheat trains over two different time scales. Within this approach, dynamic parameter estimation problems are solved at frequent intervals to update the model parameters and cope with variability and uncertainty, while predictive first principle models are used to optimize the performance of the network over a future horizon. Applying this multi-loop optimization approach to a case study of a real refinery demonstrates the importance of considering process variability on deciding about optimal fouling mitigation approaches. Uncertainty and variability have been ignored in all previous model based fouling mitigation strategies, and this novel multi-loop NMPC approach offers a solution to it so that the economic savings are enhanced.
In conclusion, the models and optimization algorithms developed in this thesis have the potential to reduce the operating cost and carbon emission of refining operations by mitigating fouling. They are based on accurate models and deterministic optimization that overcome the limitations of previous applications such as poor predictability, ignoring variability and dynamics, ignoring interactions in the system, and using inappropriate tools for decision making.