In order to fully extract the nonlinear and non-stationary characteristics of the dam deformation monitoring data, deeply mine the topological relationship between the preceding and following information, effectively improve the prediction accuracy, a bidirectional long short-term memory (LSTM) neural network dam deformation prediction model based on quadratic modal decomposition and dung beetle optimization algorithm is proposed. The model introduces a quadratic modal decomposition that combines adaptive noise-complete integrated empirical modal decomposition and variational modal decomposition to preprocess the data, which effectively reduces the negative influence of high-frequency non-stationary components on the prediction accuracy, and utilizes dung-beetle optimization algorithm to search for the optimal hyper-parameters of bidirectional LSTM neural networks to deeply explore the effective information of the deformation data of the dams. Taking a hydropower dam as an example, the prediction results of the model are compared with those of several commonly used models, and the experimental results show that the model can effectively exploit the complex nonlinear features of dam deformation data, and its prediction accuracy is significantly better than its counterparts, and the feasibility and superiority of the model in the prediction of dam deformation is verified.