In order to solve the problem that traditional dam outlier identification methods mostly rely on single measuring point models and fail to fully consider the spatio-temporal correlation characteristics of deformation between measuring points, which easily leads to misdiagnosis of outliers, a method for identifying concrete dam deformation outliers based on spatio-temporal clustering and deep learning is proposed. This method utilizes the spatio-temporal correlations of deformations between measurement points to perform spatio-temporal clustering and partitioning of the deformation data from the measurement points of concrete dams. Based on the new honey badger algorithm (HBA) and the bidirectional long short-term memory (BiLSTM) neural networks, the HBA-BiLSTM deformation prediction model is established. Deformation outliers in concrete dams are identified based on the deformation values output by the established model and outlier discrimination indices. The results of case validation show that this method has higher accuracy than traditional outlier identification methods.