This research proposes a dam deformation prediction model based on clustering partitioning and Bidirectional Long Short-Term Memory (BiLSTM) networks to address the limitations of traditional monitoring models in characterizing the distribution characteristics of deformation zones in concrete gravity dams. The model takes into account the intrinsic correlations among monitoring points and achieves more comprehensive deformation monitoring by integrating multiple feature information. Firstly, the improved K-Shape algorithm, which takes into account the time series features and spatial coordinate relationships, is used to cluster and partition the spatial measurement points to better capture the spatial distribution characteristics of the deformation region. Following that, the model hyperparameters undergo iterative optimization using the ZOA optimization algorithm to enhance overall model performance. Finally, a ZOA-BiLSTM modelling process incorporating the correlation characteristics of multiple measurement points is proposed. After validation by engineering examples, the clustering results coincide with the spatial distribution characteristics of dam deformation. Meanwhile, the prediction model has high accuracy and robustness, and predicts the dam deformation from the multi-measurement point correlation dimension, which provides a new and effective method to monitor the overall safety state of the dam.