Abstract: To address data loss in concrete dam deformation monitoring caused by equipment failure, data transmission distortion, human error, and environmental interference, this study proposes a multipoint missing data imputation model that integrates Fast Dynamic Time Warping (FastDTW) with a Bayesian-optimized Long Short-Term Memory network (BO-LSTM). The approach first employs FastDTW to identify morphological patterns and spatiotemporal correlations across multiple time series, thereby selecting highly similar reference points for the target location. Subsequently, Bayesian optimization is applied to automatically tune the hyperparameters of the LSTM network, enhancing its capacity to capture complex temporal dynamics and enabling accurate imputation and prediction of missing values. Validation using real-world monitoring data from a concrete dam demonstrates that the FastDTW-BO-LSTM model achieves consistently strong performance and robust generalization across missing rates ranging from 5% to 25%. Notably, at a 15% missing rate, the model significantly outperforms benchmark methods, achieving a coefficient of determination (R²) of 0.980, a mean absolute error (MAE) of 0.109 mm, a mean absolute percentage error (MAPE) of 0.61%, and a mean squared error (MSE) of 0.05 mm. By improving both the accuracy of data reconstruction and the reliability of deformation forecasting, the proposed model provides a high-quality data foundation essential for ensuring the safe and effective operation of dam infrastructure.