Neural network and principal component analysis for modeling of hydrogen adsorption isotherms on KOH activated pitch-based carbons containing different heteroatoms

Abstract

A three-layer feed-forward artificial neural network (ANN) was constructed and tested to model the equilibrium data of hydrogen onto activated carbons containing different heteroatoms. The properties of the activated carbons and the experimental conditions are used as inputs to predict the corresponding hydrogen uptake at equilibrium conditions. The statistical validity of activated carbon properties in discriminating the adsorbent type was carefully studied and validated. The constructed ANN was also found to be precise in modeling the hydrogen adsorption isotherms for all inputs during the training process. The trained network successfully simulates the hydrogen sorption isotherm for the new inputs, which are kept unaware of the neural network during the training process, thus showing its applicability to determine the sorption isotherms for any operating conditions under the studied limits. The absolute percentage deviation between the experimental and predicted data during the training and testing process was observed to be less than 5% for most of the input conditions.