Abstract: The longshort-term memory(LSTM) algorithm is used for model training based on the field monitoring data of a metro foundation pit to accurately predict the horizontal deformation of diaphragm walls of foundation pits. The model predicts the horizontal deformation of the diaphragm wall of the foundation pit under two conditions: excavation and postfoundation slab-pouring, and the predicted results are compared with those obtained using other models. The results show that the prediction accuracy of LSTM model is higher compared with the BP prediction and gray prediction models. The additional prediction based on multiple measuring points and working conditions indicates the recommended model′s stability and reliability.

Key words: foundation pit, deformation prediction, monitoring data, longshort-term memory algorithm