Abstract We introduce variational mode decomposition(VMD) and bidirectional long short-term memory (BiLSTM) neural network for dam deformation prediction research. Firstly, we use VMD to reduce the influence of nonlinearity and non-smoothness of the original dam data on the prediction results; secondly, we combine the hunter-prey optimizer(HPO) with the parameter optimization of BiLSTM to construct a dam deformation prediction model based on VMD-HPO-BiLSTM; finally, we compare the results of this model with LSTM, BiLSTM, and VMD-BiLSTM models using a hydroelectric dam as an example. The experimental results show that the VMD-HPO-BiLSTM model are 0.446 mm, 0.264 mm, and 18.593% in the three accuracy evaluation indexes of RMSE, MAE, and MAPE, respectively, which are better than other three models and has highest prediction accuracy.