Accurate　and　reliable　prediction　of　renewable　energy　is　critical　to　the　operation　and　optimization　of　resources　in　cloud　data　centers.　It　is　also　vital　to　reduce　energy　cost　and　harmful　gas　emission.　However,　it　is　highly　challenging　to　achieve　it　due　to　unstable　characteristics　of　renewable　energy.　Traditional　prediction　methods　are　mainly　time　series　forecasting　ones,　and　their　prediction　accuracy　is　unsatisfactory　since　they　ignore　spatial　dependence　in　wind　speed　data.　This　work　proposes　a　spatio-temporal　prediction　method　to　predict　the　wind　speed　data.　It　adopts　a　Savitzky-Golay　filter　to　smooth　the　wind　speed　data　to　reduce　the　noise　interference.　It　learns　the　spatial　dependence　through　a　graph　convolutional　network,　and　adopts　a　gated　recurrent　unit　to　extract　temporal　dependence　of　the　wind　speed　data.　In　this　way,　this　method　effectively　removes　the　noise　and　obtains　temporal　and　spatial　features　of　the　wind　speed　data,　thereby　achieving　better　prediction　accuracy.　Experimental　results　demonstrate　that　the　proposed　approach　outperforms　other　baseline　peers　by　using　real-world　datasets.