Predicting variation of multipoint earth pressure in sealed chambers of shield tunneling machines based on hybrid deep learning

Abstract

In order to avoid safety accidents caused by the unbalance of earth pressure in the chamber during the tunneling process of shield tunneling machine, it is very crucial to make a scientific and accurate dynamic prediction of earth pressure change. Therefore, a hybrid deep learning model is built by using discrete wavelet transform (DWT), one-dimensional convolution neural network (1DCNN) and long-short term memory (LSTM), implementing the scientific prediction of multipoint earth pressure variation trend in this paper. The DWT is introduced to reduce noise in training data, reduce training cost, and improve training accuracy; the 1DCNN is employed to extract features quickly and reduce the training time of LSTM; in order to further improve the prediction performance of the model, the Ranger optimizer is used to optimize the LSTM, which not only improves the prediction accuracy, but also makes the training process of the prediction model more stable. Finally, the validity of the model is verified based on the actual construction data. The results show that the overall and local earth pressure variation trends in the sealed chamber are clearly visible, which can provide decision-making basis for the shield machine to realize automatic, intelligent and safe excavation construction.