Prediction of hydrogen storage in dibenzyltoluene empowered with machine learning

Abstract

Hydrogen storage using liquid organic hydrogen carriers (LOHCs) is a promising method. The data sets for hydrogen storage using dibenzyltoluene (DBT) are considered in this study. The important input parameters to predict the hydrogen storage in DBT are temperature, pressure, stirring speed, catalyst dosage, and amount of DBT. In this manuscript, Hydrogen Storage Prediction System Empowered with Machine Learning (HSPSML) is proposed. The three different Artificial Neural Network (ANN) approaches such as Levenberg-Marquardt, Bayesian Regularization, and Scaled Conjugate Gradient are chosen to predict the hydrogen storage capacities and their results are compared to indicate the optimal approach. The data sets are classified into two classes i.e., low and high. The overall accuracy of the Bayesian Regularization (BR) and Scaled Conjugate Gradient (SCG) are 98.70 % whereas it is 94.87 % for the Levenberg-Marquardt (LM) approach. The accuracy of the LM approach is lower due to the high miss clarification rate of 12.8 % of the low class. The low class accuracy is 100 % in the other two approaches which resulted in the higher overall accuracy of these methods. Therefore, the BR and SCG are found to be the optimal approaches to predicting hydrogen storage capacities.