Load　forecasting　can　be　used　to　optimize　the　operation　of　the　energy　management　system　and　reduce　the　cost　of　energy　consumption.　In　this　paper,　we　implement　an　energy　management　system　in　the　office　building　of　Fujian　Huatuo　Automation　Technology　Company.　The　smart　meters　monitor　the　energy　consumption　of　the　building,　and　the　smart　meter　data　are　transmitted　to　the　cloud　server　for　load　forecasting.　To　improve　the　precision　of　load　forecasting,　we　adopt　the　gradient　boosting　decision　tree　(GBDT)　to　process　the　data,　and　study　the　best　combination　of　features.　The　smart　meter　data　are　used　to　test　the　performances　of　the　proposed　load　forecasting　approach,　and　the　results　show　that　the　proposed　approach　has　better　performance　than　traditional　methods.　©　2019　IEEE.