Abstract: This paper proposed a method for predicting landslide deformation rates using a dynamic serial PSO-BiLSTM approach, aiming to overcome the limitation such as insufficient accuracy and low computational efficiency found in existing methods. Initially, the deformation rate of landslides is captured through a dynamic sliding window technique, and the resulting sequence is decomposed using Ensemble Empirical Mode Decomposition (EEMD) to extract trend and periodic components. Subsequently, the deformation rate prediction sequences of trend and periodic components were obtained through polynomial fitting and a periodic component of PSO-BiLSTM network, respectively. After several cycles that produce residual deformation rate sequences, these are integrated with the initial prediction sequences to establish a comprehensive PSO-BiLSTM prediction network that yields the total predicted deformation rate. The method was validated with a landslide monitoring case in Sichuan Province, achieving a MAE of 0.28, a MAPE of 5.41%, an RMSE of 0.57, and an R² of 0.98, with a computation time of 380.22 seconds, thus ensuring high accuracy and computational efficiency.