Generation of Linear and Nonlinear Based Surrogate Models for CO2 Utilization Pathways

Abstract

Climate change from global warming, driven by greenhouse gases (GHG), urgently needs action to protect future generations. CO2 is the primary target for emission reduction. Carbon Capture and Utilization (CCU) is a promising strategy, with methanol, ammonia, and urea production from captured CO2 and renewable hydrogen offering significant potential to mitigate emissions and create valuable products. While there is extensive research dedicated to this strategy by industry, scientists and policy-makers, it requires effective methods for optimizing big data prediction and reduce production time for analysis of modelling especially since the industry is transitioning into the era of the fourth industrial revolution (Industry 4.0). Traditional methods for process modelling in Chemical engineering involve first-principal models that are accurate in simulating complex equations but are highly computationally demanding and are not easily scalable to varying process parameters leading to inefficiencies in the early design stages and enterprise wide optimization. The current study involves the generation of linear and non-linear surrogate models of three processes involved in CO2 utilization: Methanol, Ammonia and subsequent urea production. To overcome the complexity and computational issues, such models will greatly increase efficiency in the early design stage and enterprise wide optimization of eco parks for CO2 utilization involving such plants. Data driven surrogate models were developed for these processes involving production of urea from captured CO2 and renewable ammonia (NH3) which is produced from renewable hydrogen (H2). Renewable methanol process also considers the same methods. Response surface methodology (RSM) was the chosen method of surrogate modelling and various input factors including feed flow, pressure and temperature were used to predict selected response variables such as product flow and CO2 utilization. The validation of the models is conducted through tests such as coefficient of determination (R2), Root mean squared error (RMSE) and crossplots. The results indicated great potential for the models with high R2 values of above 75% for the linear methanol models and above 95% for the non-linear ones. For ammonia, the linear model had R2 values of 94.73% and for urea, a step-wise regression produced models with above 80% R2. The RMSE for all final models for the three processes at testing was within 18% deviation of the data range of the actual response from simulation data. Thus the proposed surrogate models offer a significant reduction in computational complexity compared to the simulation models while providing sufficient accuracy offering a more efficient approach for early-stage design and analysis of CO2 utilization plants and for use in enterprise wide optimization cases of CO2 utilization eco-parks.

Date of Award	26 Aug 2024
Original language	American English
Supervisor	Elkamel (Supervisor)