The　deformation　prediction　of　a　concrete　dam　is　important　to　its　safe　operation.　To　solve　the　problem　of　low　prediction　accuracy　of　traditional　analysis　methods　resulted　from　the　difficulty　in　capturing　the　characteristics　of　long-term　sequences,　this　paper　uses　a　combination　of　Sparrow　Search　Algorithm　(SSA)　and　the　K-Harmonic　Mean　(KHM)　algorithm　to　cluster　the　monitored　values　and　capture　the　long-sequence　features.　Then,　we　use　methods　such　as　Complete　Ensemble　Empirical　Mode　Decomposition　(CEEMDAN)　to　reduce　the　noise　in　the　clustered　data,　and　a　long　short-term　memory　(LSTM)　model　to　predict　long　sequences.　The　analysis　results　show　this　clustering　method　has　a　better　capability　of　identifying　long-sequence　features.　It　removes　the　redundant　information　from　the　sequence　by　cooperating　with　the　CEEMDAN　decomposition-based　method,　and　enables　the　LSTM　model　to　better　capture　the　time-sequence　characteristics　of　dam　deformation,　thus　improving　the　prediction　accuracy　significantly.　The　proposed　method　is　good　in　accuracy　and　adaptability　and　useful　for　dam　deformation　prediction.　©　2022　Tsinghua　University　Press.　All　rights　reserved.