Supervised Machine Learning-Based Prediction of Hydrogen Storage Classes Utilizing Dibenzyltoluene as an Organic Carrier
Ahsan Ali, Muhammad Adnan Khan, Hoimyung Choi
Abstract
Dibenzyltoluene (H0-DBT), a Liquid Organic Hydrogen Carrier (LOHC), presents an attractive solution for hydrogen storage due to its enhanced safety and ability to store hydrogen in a concentrated liquid form. The utilization of machine learning proves essential for accurately predicting hydrogen storage classes in H0-DBT across diverse experimental conditions. This study focuses on the classification of hydrogen storage data into three classes, low-class, medium-class and high-class, based on the hydrogen storage capacity values. We introduce Hydrogen Storage Prediction with the Support Vector Machine (HSP-SVM) model to predict the hydrogen storage classes accurately. The performance of the proposed HSP-SVM model was investigated using various techniques, which included 5-Fold Cross Validation (5-FCV), Resubstitution Validation (RV), and Holdout Validation (HV). The accuracy of the HV approach for the low, medium, and high class was 98.5%, 97%, and 98.5%, respectively. The overall accuracy of HV approach reached 97% with a miss clarification rate of 3%, whereas 5-FCV and RV possessed an overall accuracy of 93.9% with a miss clarification rate of 6.1%. The results reveal that the HV approach is optimal for predicting the hydrogen storage classes accurately.