Time Series Modelling of Temperature Distributions in Proton Exchange Membrane Fuel Cells

ABSTRACT Proton exchange membrane (PEM) fuel cells are central to net‐zero transportation, with challenges in durability, thermal management, and long‐term monitoring that limit their large‐scale deployment. Temperature distribution within the cell serves as a key diagnostic indicator, as localized hotspots accelerate degradation of cell components. In this study, a long short‐term memory (LSTM) model is developed for predicting spatial temperature distributions in PEM fuel cells using readily measurable operating parameters and current distribution data. A 500‐h experimental dataset under the new European driving cycle is employed, comprising 72 input features (24 operating parameters and 48 current sensors) and 48 temperature sensor outputs across a 50 cm ² active area. Bayesian optimization is applied for hyperparameter tuning, yielding an LSTM configuration that achieves high predictive accuracy ( R ² > 0.91) when trained on only 100 h (20%) of the dataset under the investigated hyperparameter settings. Sensor‐wise evaluation shows minimal error for central regions with uniform heat transfer, while higher deviations occur near edges due to local non‐uniformities. This suggests the need for targeted calibration and feature engineering for edge regions to further enhance predictive reliability. Overall, the framework demonstrates a computationally efficient, data‐driven pathway for predictive monitoring and diagnostics in PEM fuel cells.