Machine learning modelling and optimization for metal hydride hydrogen storage systems

Abstract

Solid-state storage is a promising way to store hydrogen due to its high energy density. However, the development of a solid-state storage system is a complex problem due to various parameters affecting the systems. Several numerical models were developed in the past to analyse the behaviour of these systems but solving these numerical models required a lot of time and expertise. In this study, various machine learning models were developed by considering the most sensitive parameters obtained by a detailed literature review which affect the performance of metal hydride systems using LaNi₅ as the storage material. The data required for training of these models were obtained by using a validated mathematical model solved using COMSOL software. The performance of all the developed machine learning models was compared and the most accurate model i.e., an artificial neural network was selected for further study. The development of this machine learning model helps designers and experts to evaluate the reaction fraction and bed temperature directly and quickly without undergoing detailed mathematical modelling which saves a lot of time and resources. Further, a new method was described in which the genetic algorithm was utilized to optimize the developed ANN model based on the application. The applicability of the developed method was then shown by optimizing various parameters at different time instances for the development of metal hydride hydrogen storage for vehicular applications.