Surrogate-based process optimization: A case study on simple natural gas processing plant

Natural gas is a vital component of the world's energy supply. Therefore, optimizing natural gas processes is important to facilitate more efficient and less costly operations. Optimization which relies exclusively on simulation is impractical, due to the enormous computational cost of complex simulations. Alternatively, surrogate models can be constructed from and used in lieu of actual simulation models. The proposed work aims to develop a black-box surrogate-based optimization model that represents the natural gas treatment process which consist of separation, sweetening and dehydration units. A combination of simulation, experimental design, data analytics and mathematical programming was utilized. The problem size was reduced by using a regression model of the system's behavior. The regression model was based on the simulation and experimental design. This simplified model was then used in a mathematical programming model minimizing a cost function. The methodology is suitable for synthesis of large-scale complex problems with fewer degrees of freedom. Linear regression along with polynomial feature transformation machine learning methods were used to generate the surrogate model. A case study was developed to investigate the impact of increasing H2S composition in the feed gas. It was observed that the feed pressure had the highest influence among the parameters.