Power　grid　is　often　destroyed　due　to　frequent　typhoon　disasters　in　coastal　area　of　China,　when　making　urgent　repair　on　it,　the　guarantee　of　emergency　supplies　of　power　grid　became　vital.　Currently,　the　sample　data　collected　by　the　power　grid　company　for　typhoon　events　is　relatively　few,　while　typhoons　events　continue　to　occur,　the　data　could　be　constantly　obtained　along　with　time　series,　but　the　traditional　predictive　methods　for　emergency　supplies　quantity　is　offline　mode　mostly.　In　this　paper,　we　come　up　with　an　Online-RVM　model　to　predict　emergency　supplies　quantity　by　extract　feature　of　typhoon　event.　In　this　model,　we　first　use　micro-genetic　algorithm　to　find　out　the　optimal　RVM　kernel　function　parameter,　and　introduce　fast　marginal　likelihood　algorithm　to　speed　up　the　model　update.　It　turned　out　that　the　prediction　model　we　came　up　with　gain　good　performance　in　small　sample　training　data　compared　with　classic　predictive　methods　for　emergency　supplies,　and　it　could　has　a　quick　and　accurate　prediction　for　emergency　supplies　quantity　in　next　typhoon　event.　©　2016　IEEE.