As　an　important　component　of　a　proton　exchange　membrane　fuel　cell,　the　structure　of　gas　diffusion　layer　affects　the　transmission　of　materials,　thus　affecting　the　heat　and　current　density.　However,　the　current　researches　on　gas　diffusion　layer　are　mostly　about　homogeneous　structure,　and　additionally　it　is　unknown　to　the　performance　influenced　by　the　structure　parameters.　Firstly,　a　gas　diffusion　layer　(GDL)　model　reflecting　complex　structure　is　constructed　based　on　a　random　parameter　method,　and　a　micro-element　model　of　fuel　cell　is　established.　Secondly,　a　Latin　hypercube　sampling　is　used　to　obtain　sample　points　and　an　extreme　learning　machine　(ELM)　model　is　formulated　to　fit　simulation　data.　The　ELM　model　is　verified　by　comparing　with　the　actual　results.　Finally,　the　micro-element　model　performance　affected　by　the　structure　variables　is　analyzed,　and　the　optimal　structural　variables　are　obtained.　The　results　show　that　the　ELM　model　can　accurately　predict　the　fuel　cell　micro-element　model　performance　and　its　prediction　accuracy　is　better　than　that　of　the　back　propagation　(BP)　neural　network　model.　The　multi-objective　optimal　algorithm　model　can　obtain　the　optimal　GDL　structure　parameters　corresponding　to　the　minimum　temperature,　maximum　current　density,　and　good　oxygen　concentration　uniformity.