Ground　settlement　is　a　critical　issue　in　underground　construction.　Ground　settlement　prediction　is　important　to　identify　serious　damage　to　adjacent　structures　caused　by　settlement　exceed　the　standard.　However,　conventional　methods　have　some　limitations　due　to　nonlinearity　and　non-stationarity　of　settlement　data.　This　paper　proposes　a　hybrid　model　based　on　empirical　model　decomposition　(EMD)　and　relevance　vector　machine(RVM)　regression　optimized　by　particle　swarm　optimization　(PSO)　designated　as　EMD-RVM.　EMD　is　used　to　decompose　the　ground　measured　settlement　time　series　into　several　stationary　components　called　intrinsic　mode　functions　(IMFs).　Then,　SVR　is　applied　to　predict　the　components　independently.　At　last,　the　expected　prediction　values　are　the　sum　of　all　components　prediction　value　at　the　same　time.　In　order　to　validate　the　performance　of　the　proposed　method.,　the　signal　support　vector　regression(SVR),　signal　relevance　vector　machine(RVM)　regression,　relevance　vector　machine　based　on　wavelet　transform(WT-RVM)　are　used　to　be　compared　with　root　mean　square　error　(RMSE)　and　mean　absolute　percentage　Error(MAPE)　are　used　to　evaluate　these　models.　The　evaluation　results　indicate　the　method　proposed　is　effective　and　practical.