To enhance the accuracy and generalization capability of concrete dam deformation prediction models, fully utilize the topological relationships of temporal data in long time series, and consider various uncertainties, an improved Transformer (ITransformer) model and bidirectional long short-term memory neural network (BiLSTM) were integrated with the Kepler optimization algorithm (KOA) to construct the KOA-ITransformer-BiLSTM concrete dam deformation prediction model, namely KIB model. Additionally, to address uncertainties such as measurement errors, the quantile regression (QR) method was introduced, establishing the QR-KOA-ITransformer-BiLSTM concrete dam deformation interval prediction model, namely QKIB model. Experimental results demonstrate that the KIB model significantly improves prediction accuracy and iterative efficiency, reducing the mean absolute error, mean absolute percentage error, mean squared error, and root mean squared error by 77.89%, 90.37%, 93.08%, and 73.69%, respectively, compared to the LSTM models, while increasing the coefficient of determination R 2 by 23.02%. The QKIB model exhibits superior stability and accuracy, with the comprehensive evaluation metric at the 90% confidence level decreasing from 1.387 36 (QR-LSTM), 0.786 48 (GRU), and 0.543 42 (QR-BiLSTM) to 0.241 95.