Blockchain　has　been　widely　used　to　solve　data　privacy　and　security　issues　in　edge　computing　scenarios.　However,　the　blockchain　based　on　edge　computing　still　has　some　performance　problems,　such　as　insufficient　scalability,　difficulty　in　balancing　security　and　edge　device　power　consumption,　and　inability　to　simultaneously　meet　low　latency,　high　throughput,　high　security　and　privacy　issues,　etc.　In　order　to　solve　these　problems,　this　paper　proposes　a　generally　improved　Byzantine　consensus　mechanism　based　on　the　K-medoids　clustering　algorithm　-　FIBFT.　Considering　the　different　performance　characteristics　of　each　node　in　the　network,　the　node＇s　state　is　first　abstracted　into　a　multi-dimensional　state　space　containing　eigenvalues,　and　then　the　nodes　are　divided　into　subnets　by　the　efficient　K-medoids　clustering　algorithm.　Each　subnet　uses　a　Byzantine　consensus　mechanism　based　on　arbitration　for　consensus　and　data　interaction,　and　the　consensus　data　could　be　exchanged　between　the　subnets　without　interfering　with　the　consensus　process.　The　research　results　show　that　FIBFT　has　better　scalability　and　throughput　while　ensuring　high　security　compared　with　the　traditional　Byzantine　consensus　algorithm.