Hydrogen can be produced from a range of renewable sources (solar or biomass). Often cost effective production of hydrogen does not align with hydrogen demand and thus it must be stored for later use to produce power for the grid. One promising method is a Proton Exchange Membrane fuel cells (PEMFC) which produce electricity and water as products from hydrogen and air.

However, PEMFC’s are complicated and require optimized control strategies to ensure their longevity and performance under a wide range of ambient conditions and varying loads. Machine Learning-based PEMFC control development methodologies can help provide cost effective and robust control to quickly accelerate market penetration of these clean alternative energy systems.

In this project, the control of PEM fuel cells stacks will be investigated to optimize the efficient production of electricity. To accomplish this task, Machine Learning (ML) is used in three areas: deep learning based system modeling, Reinforcement Learning (RL) and machine learning based model predictive control (MLMPC). The goal of this work is to improve the performance and efficiency of fuel cell (FC) systems by applying cutting edge developments in ML control. A FC stack is a complex energy conversion system that requires precise control of the flow of hydrogen and oxygen (air) both the pressures and humidity along with the stack temperature have a significant influence on the cell performance (output power) and lifespan (cell degradation).