Real-time prediction of shield moving trajectory during tunnelling using GRU deep neural network

This paper establishes an intelligent framework for real-time prediction of trajectory deviations in the process of earth pressure balance (EPB) tunnelling. A hybrid model was developed which integrates principal component analysis (PCA) and a gated recurrent unit (GRU). PCA was adopted to mine the interrelated input parameters and reduce the accompanying data noise. A scroll window mode was implemented in the GRU to predict the shield movement in real time. The proposed PCA–GRU model was implemented and validated through a case study of the Guang-Fo intercity railway in Guangzhou, China. Another three machine learning models were also used for comparison. The results revealed that the proposed model predicted the shield moving trajectory with higher precision than other models. The implications for trajectory regulation were discussed using field data. The proposed prediction framework represents a promising solution for real-time prediction of the shield moving trajectory in EPB tunnelling.