Supervised Machine Learning-Based Prediction of Hydrogen Storage Classes Utilizing Dibenzyltoluene as an Organic...

Area of Science:

• Chemical Engineering
• Materials Science
• Data Science

Background:

• Liquid Organic Hydrogen Carriers (LOHC) like Dibenzyltoluene (H0-DBT) offer safe, concentrated hydrogen storage.
• Accurate prediction of hydrogen storage capacity is crucial for LOHC applications.
• Machine learning (ML) is increasingly vital for complex material property predictions.

Purpose of the Study:

• To classify hydrogen storage data into low, medium, and high classes based on capacity.
• To develop and evaluate a machine learning model for predicting these hydrogen storage classes in H0-DBT.
• To determine the optimal validation technique for accurate hydrogen storage predictions.

Main Methods:

• Development of a Support Vector Machine (SVM) model, termed HSP-SVM, for hydrogen storage classification.
• Performance evaluation using 5-Fold Cross Validation (5-FCV), Resubstitution Validation (RV), and Holdout Validation (HV).
• Analysis of classification accuracy and miss-clarification rates for each validation method.

Main Results:

• The Holdout Validation (HV) approach demonstrated superior performance.
• HV achieved high class-specific accuracies: 98.5% (low), 97% (medium), and 98.5% (high).
• HV yielded an overall accuracy of 97% with a 3% miss-clarification rate, outperforming 5-FCV and RV (93.9% accuracy, 6.1% miss-clarification).

Conclusions:

• The Holdout Validation (HV) approach is the optimal method for accurately predicting hydrogen storage classes in H0-DBT.
• The HSP-SVM model, validated with HV, provides a reliable tool for assessing hydrogen storage potential in LOHCs.
• Accurate ML-driven predictions enhance the development and application of LOHC technologies for hydrogen energy systems.