Surrogate modeling application for process system emissions assessment: improving computational performances for plantwide estimations

During the last decade, data driven modeling has gained a role of major interest all over the engineering fields mainly due to the need of higher computational power or, inversely, less computational demanding models for applications such as optimization, simulation, scheduling and control. Relevant contributions of process systems surrogate modeling as a support for operation optimization were already proved in literature with a reduction of the overall computational time by two orders of magnitude with respect to conventional simulations. In this research work a biogas-to-methanol plant case study is used to assess the total energy consumption and estimate the related emissions. Therefore, a modeling phase carried out via Response Surface Methodology is set up in order to obtain the analytical function that allows to estimate the equivalent CO2 emissions over an extended range of operating conditions representing a wide interval of biomass feed composition. The study has been performed over a wide independent variables domain as well as for different sample sizes in order to compare the computational performances and the accuracy of the obtained models accordingly. The computational time was reduced by two orders of magnitude with a mean relative error lower than 1%. Given the quality of the results, this approach could be further exploited for other system variables and processes including highly non-ideal behaviour of mixtures to be treated. Furthermore, more complex sampling and different surrogate modeling strategies could be tested in order to check if even higher computational effectiveness and model accuracy could be obtained in the process systems domain.