Surrogate-Based Optimization of the OPEX of a Modular Plant for Biogas Conversion to Methanol Using the MADS Algorithm

The present work studies the potential of surrogate models for the global optimization of complex chemical processes. In particular, a modular plant for the conversion of biogas to methanol is considered. The Aspen HYSYS simulation of this plant was run 480 times, which ensured the even distribution of points in the input space. The evenness of this design of experiments was evaluated using a discrepancy measurement called the Mixture Discrepancy. With the simulation data, some of the most widely used surrogate models such as regression models and the Kriging Gaussian process were trained. The most accurate model for the prediction of each output variable was selected and used for the optimization of the OPEX. The optimization complemented the trained surrogate models with the Mesh Adaptive Direct Search (MADS) algorithm. For this purpose, the openaccess computational implementation of the MADS algorithm called NOMAD was used. With the surrogate-based optimization, the computational times were reduced an 88% with respect to the simulation-based optimization. In addition, the accuracy of the surrogate model was paramount, as an average 0.75% prediction error was found. Consequently, the models proved sufficient for optimizing the studied process, resulting in a 22.2% reduction in the OPEX.