Application of extreme learning machine for prediction of aqueous solubility of carbon dioxide

Abstract

Solubility of CO2 in brine is one of the contributing trapping mechanisms by which the injected CO2 is sequestrated in aquifers. In the literature, the solubility data on low salinity range are scarce. Thus, in the current study, the CO2 solubility was experimentally obtained in the NaCl brines of low salinity (0–1.5 wt%) at temperature of 333–373 K and pressures up to 280 MPa through the potentiometric titration methods. The short-term, multistep ahead predictive models of aqueous solubility of carbon dioxide were created. The models were developed using a novel method based on the extreme learning machine (ELM). Estimation and prediction results of the ELM model were compared with the genetic programming (GP) and artificial neural networks (ANNs) models. The results revealed enhancement of the predictive accuracy and generalization capability through the ELM method in comparison with the GP and ANN. Moreover, the results indicate that the developed ELM models can be used with confidence for further work on formulating a novel model predictive strategy for the aqueous solubility of carbon dioxide. The experimental results hinted that the current algorithm can present good generalization performance in the majority of cases. Moreover, in comparison with the conventional well-known learning algorithms, it can learn thousands of times faster. In conclusion, it is conclusively found that application of the ELM is particularly promising as an alternative method to estimate the aqueous solubility of carbon dioxide.