In　the　case　of　emergencies,　it　is　difficult　to　subjectively　estimate　the　demand　for　specific　types　of　materials.　The　uncertainty　of　demand　will　affect　the　distribution　of　materials　and　the　implementation　of　relief　work,　resulting　in　serious　casualties　and　economic　losses.　This　paper　aims　to　improve　the　prediction　accuracy　of　the　demand　for　emergency　supplies　and　provide　data　support　for　the　allocation　strategy.　Firstly,　based　on　data-driven　thinking,　the　projected　number　of　people　in　demand　was　obtained　by　LSTM　neural　network.　Then,　we　defined　the　allocation　lead　time　according　to　the　demand　characteristics　of　different　phases　of　emergency　events　and　analyzed　the　relationship　between　the　amount　of　allocation　and　the　number　of　people　in　demand.　Following　that,　combined　with　safety　stock　theory,　the　LSTM-Cubic　model　for　emergency　supplies　was　established　to　forecast　the　demand　for　emergency　supplies.　Finally,　the　validity　of　the　proposed　model　was　verified　by　using　the　outbreak　of　the　COVID-19　epidemic　in　Wuhan　at　the　end　of　2019　as　a　case　study.　The　results　show　that　the　proposed　LSTM-Cubic　model　has　higher　accuracy　compared　with　other　models,　and　is　applicable　to　the　prediction　of　sudden　demand　at　the　early　stage　of　an　emergency.　©　11th　International　Symposium　on　Project　Management,　ISPM　2023.　All　rights　reserved.