Machine Learning for Guiding High-Temperature PEM Fuel Cells with Greater Power Density

Abstract

High-temperature polymer electrolyte membrane fuel cells (HT-PEMFCs) are enticing energy conversion technologies because they use low-cost hydrogen generated from methane and have simple water and heat management. However, proliferation of this technology requires improvement in power density. Here, we show that Machine Learning (ML) tools can help guide activities for improving HT-PEMFC power density because these tools quickly and efficiently explore large search spaces. The ML scheme relied on a 0-D, semi-empirical model of HT-PEMFC polarization behavior and a data analysis framework. Existing data sets underwent support vector regression analysis using a radial basis function kernel. Additionally, the 0-D, semi-empirical HT-PEMFC model was substantiated by polarization data; and synthetic data generated from this model was the subject to dimension reduction and density-based clustering. From these analyses, pathways were revealed to surpass 1 W cm-2 in HT-PEMFCs with oxygen as the oxidant and CO containing hydrogen.