The　Industrial　Internet　of　Things　(IIoT)　accommodates　a　huge　number　of　heterogeneous　devices　to　bring　vast　services　under　a　distributed　computing　scenarios.　Most　productive　services　in　IIoT　are　closely　related　to　production　control　and　require　distributed　network　support　with　low　delay.　However,　the　resource　reservation　based　on　gross　traffic　prediction　ignores　the　importance　of　productive　services　and　treats　them　as　ordinary　services,　so　it　is　difficult　to　provide　stable　low　delay　support　for　large　amounts　of　productive　service　requests.　For　many　productions,　unexpected　communication　delays　are　unacceptable,　and　the　delay　may　lead　to　serious　production　accidents　causing　great　losses,　especially　when　the　productive　service　is　security　related.　In　this　article,　we　propose　a　brain-like　productive　service　provisioning　scheme　with　federated　learning　(BrainIoT)　for　IIoT.　The　BrainIoT　scheme　is　composed　of　three　algorithms,　including　industrial　knowledge　graph-based　relation　mining,　federated　learning-based　service　prediction,　and　globally　optimized　resource　reservation.　BrainIoT　combines　production　information　into　network　optimization,　and　utilizes　the　interfactory　and　intrafactory　relations　to　enhance　the　accuracy　of　service　prediction.　The　globally　optimized　resource　reservation　algorithm　suitably　reserves　resources　for　predicted　services　considering　various　resources.　The　numerical　results　show　that　the　BrainIoT　scheme　utilizes　interfactory　relation　and　intrafactory　relation　to　make　an　accurate　service　prediction,　which　achieves　96%　accuracy,　and　improves　the　quality　of　service.