Robust group intelligent models for predicting hydrogen density and viscosity: Implication for hydrogen production, transportation, and storage

Abstract

Background

Accurate hydrogen density and viscosity determinations are crucial for optimizing processes, enhancing energy efficiency, and ensuring safety in fuel cells and storage.

Methods

In this study, we propose new robust machine learning (ML) models using decades of data to predict hydrogen density and viscosity across various pressures and temperatures. The ML-based viscosity models were developed using 1063 measurements under pressure and temperature ranges of 0.006–216.443 MPa and 14–2128 K, respectively, while the density models were implemented using 368 data points covering pressure and temperature intervals of 0.098–216.443 MPa and 150–423.15 K, respectively. Our approach combines multilayer perceptron (MLP) and cascaded forward neural network (CFNN) models, integrated through the group method of data handling (GMDH), to form an advanced committee machine intelligent system (CMIS-GMDH). Additionally, new explicit expressions are implemented using multi-gene genetic programming (MGGP) to predict hydrogen density and viscosity.

Significant findings

The results demonstrated that the implemented correlations and CMIS-GMDH models offer precise predictions of the two parameters. Besides, analyses of the prediction performance exhibited that the introduced CMIS-GMDH is the most accurate paradigm by achieving small root mean square error (RMSE) values of 0.0983 and 0.1754 for density and viscosity, respectively. Furthermore, the comparison with previous studies revealed that the CMIS-GMDH models yield superior accuracy in hydrogen density and viscosity estimations. Lastly, the physical validity of the best models was investigated by carrying out thorough trend analyses.