Real-time prediction of shield moving trajectory during tunnelling

This paper presents a novel deep learning model for real-time prediction of shield moving trajectory during tunnelling. The proposed model incorporates a wavelet transform (WT) into Adam-optimised long short-term memory (LSTM) (WT-Adam-LSTM). The WT is employed to remove the irrelevant noise of data in the time and frequency domains, which allows the sequence pattern to be detected easily. The Adam algorithm is used to increase the reliability and optimise the gradient training process of the LSTM neural network for a given time series. The developed model considers the shield performance database, complex geological conditions, soil geometry, and operational parameters. A case study of a tunnel section under Bao'an International Airport was employed to verify the performance of the proposed model. A comparison with other models, i.e. recurrent neural network, LSTM, and support vector regression, was also made. The results show that WT-Adam-LSTM provides an effective solution and can achieve better results compared with other models.

Automated summary

This paper presents a novel deep learning model, WT-Adam-LSTM, for real-time prediction of shield moving trajectory during tunnelling. The model combines wavelet transform and Adam-optimised LSTM to remove noise and extract sequence patterns. It considers various factors such as shield performance database, complex geological conditions, soil geometry, and operational parameters. A case study was conducted to verify its performance and it outperformed other models like recurrent neural network, LSTM, and support vector regression.