Prediction Model of LSTM Dam Deformation Safety Monitoring Considering Variable Hysteresis

To further improve the prediction accuracy of dam deformation, fully reflecting the hysteresis of the impact of external environmental variables on dam deformation, and considering the time lag effect of variables affecting dam deformation, a modified LSTM (long short-term memory) dam deformation prediction model was proposed by using the LSTM algorithm. The input data was divided into delay variables passing through LSTM storage block and non delay variables not passing through storage block, which made the model more reasonable in physical interpretation. In order to improve the nonlinear expression ability of the prediction model, a second hidden layer was added, so that the time effect and other non delayed variables could be directly used in the last time step without complex transformation to form the subsequence required for the original input. The reliability and accuracy of the improved model were verified by calculating a specific case. The results show that the mean absolute error (MAE) and mean square error (MSE) of the improved LSTM model are 11.94% and 25.60% lower than those of the LSTM model, respectively, and have higher prediction accuracy.The positive and negative distribution range of the prediction residuals of the improved model is smaller than that of the LSTM model, and the prediction values generally change near the measured values, which can more reasonably predict the dam deformation. The prediction results of the improved LSTM model are better than those of LSTM model, which can more reasonably predict the dam deformation.