To　solve　the　problem　of　model　heterogeneity　in　federated　learning　intrusion　detection,　a　model　heterogeneous　federated　intrusion　detection　framework　based　on　knowledge　distillation　(MHFLID-KD)　is　proposed　in　this　paper.　This　framework　consists　of　two　parts:　grouping　and　knowledge　distillation　and　aggregation　in　the　dual　teacher-student　paradigm.　First,　the　server　groups　nodes　according　to　the　model　type,　and　the　models　of　the　same　type　are　assigned　to　the　same　group,　while　models　of　different　types　are　assigned　to　different　groups.　Secondly,　a　knowledge　distillation　polymerization　method　based　on　the　dual　teacher-student　paradigm　is　proposed.　The　top　two　nodes　are　selected　as　teachers　according　to　their　accuracy　in　the　knowledge　distillation　of　the　two-teacher-student　paradigm.　We　use　two　public　datasets,　NSL-KDD　and　UNSW-NB15,　in　different　heterogeneous　scenarios　to　verify　the　superiority　of　the　MHFLID-KD　framework.　The　experimental　results　show　that　this　framework　not　only　solves　the　problem　of　model　heterogeneity　in　federated　intrusion　detection　but　also　improves　the　detection　accuracy　of　nodes.　©　2022　ACM.