文献类型：期刊文献

英文题名：Unveiling compensatory flow in PEM fuel cells assisted by machine learning

作者：Wang, Bin;Pan, Weitong[1];Tian, Xinming;Tang, Longfei;Chen, Xueli[1];Wang, Fuchen

机构：[1]East China Univ Sci & Technol, Inst Clean Coal Technol, Shanghai 200237, Peoples R China; East China Univ Sci & Technol, Engn Res Ctr Resource Utilizat Carbon Containing W, Minist Educ, Shanghai 200237, Peoples R China; East China Univ Sci & Technol, State Key Lab Coal Liquefact Gasificat & Utilizat, Shanghai 200237, Peoples R China

年份：2026

卷号：420

外文期刊名：FUEL

收录：;EI(收录号：20261020226778);WOS:【SCI-EXPANDED(收录号:WOS:001710271500001)】；

基金：The research is supported by National Key Research and Develop-ment Program of China (2024YFB4006705) .

语种：英文

外文关键词：Proton exchange membrane fuel cell; Flow non-uniformity; Compensatory flow; Machine learning; Genetic algorithm

摘要：Non-uniform flow distribution-induced compensatory flow enhances the performance of scaled-up Proton Exchange Membrane (PEM) fuel cells and is therefore promising. Nevertheless, the factors governing its induction have not been comprehensively clarified. The objective of this work is to unveil the compensatory flow with the assistance of machine-learning techniques. First, a Gas Flow Channel (GFC)-Gas Diffusion Layer (GDL) flow model is established. A one-factor-at-a-time analysis is conducted to examine the effects of key geometric and operational parameters. The contributors to compensatory flow enhancement are identified, including increased pressure differences between channels, enhanced permeability, or shortened transport distances. Second, a multifactor assessment is performed. The random forest approach is employed to elucidate the feature importance. Spearman and Pearson correlation coefficients are also jointly utilized. According to the criteria of efficiently enhancing compensatory flow while avoiding elevated flow resistance, the GDL porosity is identified as the highest design priority, whereas the operational pressure has the lowest priority. Subsequently, the optimal solution is obtained using the genetic algorithm. At this stage, the compensatory flow velocity and the pressure drop are 1.666 m center dot s-1 and 2.967 & times; 104 Pa, respectively, achieving an effective trade-off. Third, an additional simulation of the flow model confirms the accuracy of the optimal solution. Furthermore, the full-scale fuel cell model simulation demonstrates that the compensatory flow enhances cell power density without incurring additional parasitic losses, thereby improving net power density from 0.875 W center dot cm- 2 to 0.905 W center dot cm- 2. The beneficial effect of the optimal solution on cell performance is verified. Finally, it is extended to a manifoldintegrated layout, demonstrating robustness under natural flow distribution. As the manifold size decreases, the FUI increases from 12.74% to 72.55%, accompanied by an enhancement in current density from 2.332 A center dot cm- 2 to 2.426 A center dot cm- 2.