Dam　deformation　is　an　important　indicator　of　the　overall　health　condition　of　a　dam.　Numerical　prediction　of　such　displacements　based　on　real　monitoring　data　is　a　common　approach　for　dam　safety　assessment.　However,　most　existing　prediction　techniques,　such　as　traditional　statistical　methods,　support　vector　machine-based　multivariate　linear　regression　models,　and　other　conventional　machine　learning　methods,　fail　to　account　for　the　dynamic　evolution　characteristics　of　dam　deformation.　These　methods　treat　the　deformation　process　as　a　static　fitting　problem,　which　leads　to　limitations　in　both　prediction　accuracy　and　generalization　capability　and　makes　long-term　forecasting　difficult.　This　study　proposes　a　deep　learning　model　ensemble　strategy　to　achieve　accurate　and　stable　long-term　predictions　of　concrete　dam　deformation　data.　The　IPSO　optimization　algorithm　is　used　to　search　for　optimal　hyperparameter　combinations　for　the　GRU　model,　which　is　difficult　to　determine　manually,　thus　enabling　the　model　to　achieve　its　best　performance.　Through　comparisons　of　the　prediction　results　from　different　models,　the　proposed　model　demonstrates　excellent　performance　in　both　prediction　accuracy　and　the　stability　of　long-term　forecasting　tasks.　This　model　holds　considerable　engineering　application　value　for　long-term　prediction　of　concrete　dam　deformation.　©　Published　under　licence　by　IOP　Publishing　Ltd.