Voltage　sag　risk　assessment　is　helpful　for　voltage　sag　prevention　and　sensitive　user　siting.　The　simulation-based　methods　have　a　large　amount　of　data　which　cannot　reflect　the　influence　of　actual　environment,　and　the　data-driven　methods　have　few　samples　with　uneven　distribution.　So　it　is　difficult　to　know　the　voltage　sag　risk　of　each　bus　in　the　whole　network.　Therefore,　a　voltage　sag　risk　assessment　method　based　on　the　fusion　of　simulated　and　measured　data　is　proposed.　First,　the　influencing　factors　of　voltage　sag　risk　are　selected　from　simulated　and　measured　data,　then　the　comprehensive　quantitative　index　of　the　influence　domain　characterizing　the　voltage　sag　propagation　characteristics　is　constructed.　Second,　the　tail　class　oversampling　and　the　head　class　undersampling　are　used　to　construct　the　simulation　source　domain　dataset,　and　the　gradient　descent　method　is　improved　by　knowledge　transfer　and　Armijo-Goldstein　criterion　to　construct　the　residual　voltage　multiple　regression　prediction　model　for　the　buses　without　measured　data.　Finally,　the　voltage　sag　risk　level　is　classified　by　combining　the　predicted　results　with　the　voltage　sag　tolerance　characteristics.　The　analysis　of　an　actual　power　grid　example　shows　that　the　accuracy　and　convergence　performance　of　the　proposed　method　are　improved　compared　with　the　existing　commonly　used　methods,　and　the　voltage　sag　risk　in　the　whole　network　can　be　evaluated　quickly　and　accurately.　©　2023　Automation　of　Electric　Power　Systems　Press.　All　rights　reserved.