A machine learning approach for modeling and optimization of a CO2 post-combustion capture unit. (15th January 2021)

• Record Type: Journal Article
• Authors: Shalaby, Abdelhamid; Elkamel, Ali; Douglas, Peter L.; Zhu, Qinqin; Zheng, Qipeng P.
• Is Part Of: Energy. Volume 215(2021)Part A
• Publication Date: 2021-01-15
• Subjects: Data analytics -- Carbon capture -- Process modeling -- Process optimization -- Machine learning -- Post-combustion; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.119113
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Reducing CO2 emissions from fossil fuel fired power plants has been a major environmental concern over the last decade. Amongst various carbon capture and storage (CCS) technologies, the utilization of solvent-based post-combustion capture (PCC), played a major role in the reduction of CO2 emissions. This paper illustrates the development of machine learning models to predict the outputs of the PCC unit. A fine tree, Matérn Gaussian process regression (GPR), rational quadratic GPR, and squared exponential GPR models were developed and compared with a feed-forward artificial neural network (ANN) model. An accuracy of up to 98% in predicting the process outputs was achieved. Furthermore, the models were utilized to determine the optimum operating conditions for the process using a sequential quadratic programming algorithm (SQP) and genetic algorithm (GA). The use of the machine learning models has proven to be very useful since the complete mechanistic model is too large, and its runtime is too long to allow for rigorous optimal solutions. The machine learning models and optimization problems were developed and solved using MATLAB. The data used in this work was obtained from simulating the process using gPROMS process builder. The inputs of the model were selected to be reboiler duty, condenser duty, reboiler pressure, flow rate, temperature, and the pressure of the flue gas. The models were able to accurately predict the outputs of the process which are the system energy requirements (SER), capture rate (CR), and the purity of condenser outlet stream (PU). Highlights: A machine learning approach of the CO2 post-combustion capture process is prepared. Simulation of the unit was first made using gPROMS and a detailed mechanistic model. The machine learning model faired well compared to the detailed simulation. With much gain in computational effort optimization studies were carried out utilizing the machine learning model. Various capture scenarios are investigated.