Use of deep learning, denoising technic and cross-correlation analysis for the prediction of the shield machine slurry pressure in mixed ground conditions

Abstract

Tunnel face stability constitutes a significant challenge for shield tunneling in urban areas. The use of a slurry pressure balanced shield machine in mixed ground conditions containing mudstone is generally disturbed by clogging, which results in noises and large fluctuations in the tunnel face pressure. These fluctuations add difficulties to the prediction of the slurry pressure. This paper proposes a denoising method to overcome this difficulty. This method is combined with variational mode decomposition and detrended fluctuation analysis. The method is coupled with cross-correlation analysis (CCA) and a long short-term memory (LSTM) network to predict the tunnel face pressure using both tunneling parameters and geological data as input. The paper proposes a predictive strategy that separates the trend component and fluctuation component from the denoised tunneling data via CCA. Two LSTM-based predictors are presented and combined for the development of a new modeling strategy. The performances of the proposed strategy are illustrated through an application to the Nanning metro. The LSTM model with the proposed strategy gave excellent results in both mudstone and round gravel grounds with an overall R2 value of 0.9974. The paper also presents a comparison of the proposed model with some traditional models as well as a discussion on the importance of input features in different ground conditions.