Intelligent prediction models based on machine learning for CO₂ capture performance by graphene oxide-based...

Area of Science:

• Materials Science
• Chemical Engineering
• Environmental Science

Background:

• Graphene oxide (GO) derived from biomass shows promise for CO2 adsorption.
• Developing efficient adsorbents is crucial for CO2 capture applications.
• Predictive modeling can accelerate the discovery of novel CO2 adsorbents.

Purpose of the Study:

• To develop a machine learning model for predicting CO2 adsorption capacity of GO-based adsorbents.
• To identify key parameters influencing CO2 uptake by GO materials.
• To minimize experimental efforts in designing efficient CO2 adsorbents.

Main Methods:

• Data extraction from 17 articles on GO-based solid sorbents.
• Utilizing specific surface area, pore volume, temperature, and pressure as input features.
• Employing seven machine learning models, including Support Vector Machine, Gradient Boosting, Random Forest, and Artificial Neural Networks (ANN) with Multilayer Perceptron (MLP) and Radial Basis Function (RBF).

Main Results:

• The ANN based on MLP demonstrated the best performance with R² > 0.99.
• Optimal MLP hyperparameters included a hidden layer size of [45 35 45 45], Adam optimizer, learning rate of 0.003, and 1971 epochs.
• Three-dimensional diagrams illustrated CO2 uptake dependency on input parameters, and MLP network characteristics were reported.

Conclusions:

• Machine learning, particularly MLP-based ANN, accurately predicts CO2 adsorption capacity for GO-based materials.
• The developed model can significantly reduce experimental work in selecting efficient CO2 adsorbents.
• This approach supports the rational design of porous GO for effective CO2 separation and cleaner manufacturing.