Applying the wavelet neural network to estimate hydrogen dissolution in underground sodium chloride solutions

Abstract

It is possible to store large-scale hydrogen in underground sodium chloride solutions. Accurate knowledge of hydrogen-brine phase equilibrium is necessary for fully benefiting from this process and successful hydrogen storage in the underground brine. Hence, it is essential to develop a reliable method for accurately monitoring the hydrogen dissolution in brine. This study utilizes the wavelet neural network (WNN) to relate the hydrogen storage ability of brine to its main influential variables, i.e., pressure, temperature, and NaCl molality. Akaike information criterion demonstrates that a single hidden layer WNN with thirteen neurons is the most efficient topology for the given purpose. This model accurately monitors hydrogen-brine phase equilibrium with the mean squared error of 2.65 × 10−5 and regression coefficient of 0.99915. Relevancy analysis shows that temperature and pressure increase, and NaCl concertation decreases brine hydrogen storage capacity. The leverage method distinguishes 257 valid measurements and six outliers in the gathered databank.