A Machine Learning Approach for Evaluating Operating Parameters in Proton Exchange Membrane Fuel Cells Using Automotive Dynamic Test Data: Conceptual Consideration for Unmanned Aerial Vehicle Applications

Owing to their elevated energy density and rapid refuelling capability, proton exchange membrane fuel cells (PEMFCs) hold strong potential as a primary power source for fully electric ground and aerial vehicles, while also generating zero greenhouse gas emissions. Nevertheless, accurately estimating PEMFC state of health (SOH) continues to pose a major obstacle to their broader adoption in electric vehicle applications. Given that PEMFC health is directly linked to essential operating parameters, particularly voltage and power output, continuous monitoring is essential for ensuring safe, dependable vehicle operation. Motivated by this need, the present work implements sophisticated data-driven machine learning techniques to estimate PEMFC output power as a basis for health monitoring. In particular, CatBoost, Decision Tree, and Random Forest algorithms were applied to a publicly accessible dataset obtained under an automotive dynamic driving cycle. Findings indicate that all three algorithms deliver low prediction error when estimating output power. Beyond this, the study provides a thorough comparison of these three machine learning approaches and discusses at a conceptual level how the proposed technique could be incorporated into energy management and trajectory planning systems for PEMFC-driven unmanned aerial vehicles.