Optimization of a Sour Water Stripping Plant Using Surrogate Models

In this work we present a methodology for the large scale optimization of complex chemical processes. In the proposal, processes have been simulated using modular simulators, where units are classified depending on their numerical noise and the CPU time needed to converge. If the unit does not generate numerical noise, then the unit is kept in the simulator. However, if we have noisy and/or CPU time consuming units, these units are replaced by a surrogate model. In addition, some of these units can be aggregated to decrease the complexity of the model. And finally, some unit operations and constraints are included as explicit equations. As a result, we solve a hybrid simulation-optimization model formed by units in the original flowsheet, surrogate models, and explicit equations. As a case study, we perform the multiobjective optimization of a sour water stripping plant, where stripping columns are replaced by Kriging metamodels, because they have proven to be accurate and reliable, and they allow a fast interpolation. For this purpose, we simultaneously consider economic aspects, heat integration and environmental impact. Our optimization strategies ensure the convergence to a local optimum for numerical noise-free models, and a solution around the optimum inside the tolerance of the numerical noise.