Multi-objective genetic algorithm optimization with an artificial neural network for CO2/CH4 adsorption prediction in metal–organic framework. (1st October 2021)

• Record Type: Journal Article
• Authors: Yulia, Fayza; Chairina, Intan; Zulys, Agustino; Nasruddin,
• Is Part Of: Thermal science and engineering progress. Volume 25(2021)
• Publication Date: 2021-10-01
• Subjects: Adsorption -- Metal organic framework -- Artificial neural network -- Multi-objective optimization -- Genetic algorithm; Heat engineering -- Periodicals; Heat engineering; Thermodynamics; Periodicals; 621.402
• DOI: 10.1016/j.tsep.2021.100967
• Languages: English
• ISSNs: 2451-9049

Abstract

Highlights: ANN modelling provides only 1.3% error to predict gas adsorption with MOF. CO2 uptakes, heat of adsorption, and CO2 /CH4 selectivity are optimized with MOGA. Applying machine learning and MOGA optimization under industrial requirements. Abstract: The industry requirement of separating gas mixtures via adsorption techniques is rapidly being imposed, as the adsorption method is regarded as superior in terms of thermodynamic efficiency and cost. Gas mixture adsorption investigations with metal–organic frameworks (MOFs) have been conducted both in experimental and molecular simulations. Molecular simulation studies are faster in predicting CO2 adsorption performance but are difficult to conduct because they require detailed information on the characteristics of the MOF. Therefore, in this study, the separation factor of CO2 /CH4, the gas adsorption performance, and the heat of adsorption were predicted using artificial neural networks (ANNs). MOF texture properties that contribute to the performance are the input in this simulation. Operating working pressure and temperature are also inputs in this simulation. Optimization is conducted using the multiobjective genetic algorithm method to maximize the separation factor and CO2 uptake with mild heat of adsorption. Moreover, the optimal values will be determined via the technique for order of preference by similarity to the ideal solution (TOPSIS). Interestingly, the amount of CO2 adsorption, selectivity, and heat of adsorption are in satisfactory agreement with the values that are predicted by ANN with high validity regressing (R = 0.99). The output optimum point to get maximum capacity of CO2 and selectivity with mild heat of adsorption are 9.97 mmol/g, 362.92 kJ/kg, and 11.01 respectively. These results provide a basis for the use of machine learning algorithms in conjunction with multiobjective optimizations to investigate the output performance of gas adsorption under the requirements of industrial applications.