Creating a machine learning based surrogate model for modelling of a carbon capture plant

Abstract

Dagens metoder for å gjøre en parameterstudie av et karbonfangstanlegg er ofte en treg prosess som krever mye innsats og manuell input. For å gjøre denne pros- essen mindre arbeidskrevende ble et neuralt nettverk konstruert i keras/tensorflow og trent på data produsert av CO2Sim. Modellen hadde en MAPE på 3.8103%, og både responsvariablene og den økonomiske analysen gav verdier som var innen- for det man skulle forvente når man sammenlignet med verdier fra eksisterende forskning.

When conducting case studies of carbon capture plants, the current solutions that exist are both tedious and time consuming. In order to speed up this process, a neural network model can be trained on data produced by a conventional simu- lation tool, and then in turn perform the case study. The neural network model was created using CO2Sim and keras/tensorflow and had a mean average percent- age error of 3.8103%, and returned both plant specifications and economic data that are within the bounds of what one would expect when compared to existing literature.