As　knowledge　is　time-sensitive,　some　researchers　have　started　to　focus　on　dynamic　knowledge　graphs　to　provide　time-dimensioned　knowledge　content　thus　reflecting　richer　information.　But　they　have　not　yet　combined　temporal　information　at　different　granularities.　Also,　in　the　case　of　multiple　knowledge　graphs　distributed　across　different　clients,　it　is　of　interest　to　ensure　that　the　knowledge　graph　embedding　representations　are　learned　without　exposing　data　and　collaboratively.　Therefore,　in　this　paper,　we　propose　a　framework　for　multi-granularity　dynamic　knowledge　graph　embedding　in　federated　learning　(FedMDKGE),　which　allows　multiple　parties　to　interact　securely　with　temporal　information　at　different　granularities.　In　the　client,　we　present　a　multi-granularity　dynamic　knowledge　graph　embedding　model　that　improves　the　capability　of　dynamic　knowledge　graph　embedding　representation　by　focusing　on　multi-granularity　temporal　facts　from　the　perspective　of　knowledge　utilization.　On　the　server,　we　design　a　multi-granularity　aggregation　rule　to　accommodate　multi-party　information　aggregation　at　different　granularities.　Finally,　we　conduct　extensive　experiments　to　demonstrate　the　superior　performance　of　our　model.　The　results　on　these　real　datasets　show　that　FedMDKGE　considering　multi-granularity　temporal　information　performs　better　than　all　comparative　baselines　and　interconnect　information　for　multi-party　dynamic　knowledge　graph　embedding　without　exposing　data.