A binary hyperbox classifier model for hydrogen storage in magnesium (Mg) and complex hydrides

Hydrogen cannot be easily stored for energy applications. One potential solution is the storage of hydrogen within metal hydrides. The main method for determining the viability of a metal hydride for hydrogen storage is through costly and time-consuming experimentation. Machine learning provides an economical solution as it determines the association between the hydrogen storage capacity and the other properties of the material. In this thesis, a binary classifier model was developed for predicting a metal hydride’s viability for storage applications. The classifier was trained on a subset of the US Department of Energy metal hydride database using the enhanced binary hyperbox approach. This work focuses specifically on complex and Mg hydrides. The algorithm was able to generate a classifier model consisting of mechanistically plausible if/then rules that predict hydrogen storage capacity from heat of formation, operating temperature, and pressure as inputs. The model had a false positive rate of 22.0% and false negative rate of 21.1%