Prediction of CO2 solubility in water at high pressure and temperature via deep learning and response surface methodology
Zohreh Khoshraftar, Ahad Ghaemi
Abstract
In the present study, temperature of 313.15–473.15 K and pressure of 0.5–200 MPa have been developed for the CO2 solubility simulations via deep learning artificial (ANN) neural network approach and response surface methodology (RSM). Levenberg Marquardt backpropagation algorithm has been selected from MLP and compared with RBF. The number of neurons twenty and ten has been selected for the first and second hidden layers, respectively. RSM and ANN models for CO2 solubility produced mean R2 values of 0.9617 and 0.9998, respectively. The best MSE validation performance of MLP and RBF networks were 0.00294229 and 0.000190 at 6 and 150 epochs, respectively.
Topics & Concepts
SolubilityResponse surface methodologyBackpropagationArtificial neural networkArtificial intelligenceDeep learningComputer scienceMachine learningMaterials scienceChemistryPhysical chemistryPhase Equilibria and ThermodynamicsCarbon Dioxide Capture TechnologiesCO2 Sequestration and Geologic Interactions