Open Access

ARTICLE

Effective Prediction of Aging and Remaining Useful Life of Proton Exchange Membrane Fuel Cell via Kolmogorov-Arnold Network Based Gated Recurrent Unit

Wenqiang Xie^*, Xiaolong Xiao, Fangfang Zhu, Ziran Guo, Xiaoxing Lu

China State Grid Jiangsu Electric Power Co. Ltd. Research Institute, Nanjing, 211103, China

* Corresponding Author: Wenqiang Xie. Email:

(This article belongs to the Special Issue: Artificial Intelligence-Driven Collaborative Optimization of Electric Vehicle, Charging Station and Grid: Challenges and Opportunities)

Energy Engineering 2026, 123(11), 15 https://doi.org/10.32604/ee.2025.075298

Received 29 October 2025; Accepted 18 December 2025; Issue published 24 September 2026

Abstract

In the framework of the comprehensive energy transition, the proton exchange membrane fuel cell (PEMFC) powered by renewable energy emerges as a promising alternative, particularly with relevance to applications like electric vehicles (EVs), where clean and efficient power sources are crucial. However, the accurate prediction of PEMFC performance degradation poses significant challenges due to the combined effects of complex and variable aging mechanisms and operational conditions, which are especially critical in the context of EV applications where reliability and durability directly impact vehicle performance and user safety. These challenges pose notable constraints on the feasibility of its large-scale commercialization in the automotive sector. To address these issues, this study proposes an efficient prediction method, namely the Kolmogorov-Arnold network-based gated recurrent unit (GRU-KAN), aiming to improve prediction accuracy and optimize maintenance management strategies for PEMFC systems in EVs. Firstly, median filtering is employed to mitigate the interference from noise and extreme environmental factors. Furthermore, Pearson correlation analysis is utilized to screen out parameters strongly correlated with the output. Lastly, GRU-KAN is applied to predict the aging trend and remaining useful life (RUL). To validate the superiority of GRU-KAN, aging data from different operating states are employed. Simulation results demonstrate that GRU-KAN exhibits higher prediction accuracy compared to other methods. For instance, under a dynamic state and with a 50% training set, the root-mean-square error of aging prediction errors for GRU-KAN is reduced by 77.08% compared to GRU, while also providing the most accurate RUL prediction results, thereby enhancing the prognostic capability for PEMFC health management in electric vehicles.

---

Keywords

---