Research on prediction method of section convergence and deformation of shield tunnel in operation period based on deep learning

Abstract

Presently, the tunnel construction of China is changing from "mainly construction" to "equal emphasis on construction and maintenance". In this context, tunnel structure inspection, monitoring, and maintenance technologies have developed rapidly. However, the effective use and in-depth mining of massive data has always been the most difficult point in this field. Based on deep learning technology, this paper carries out in-depth mining of the multi-source data of the tunnel structure, so that the convergence and deformation of the section during the operation of the shield tunnel is predicted in a short time. Taking the Nanjing Yangtze River shield tunnel project as an example, the indicators are screened based on the fluctuations of the pair wise correlation coefficients of all monitoring indicators. And based on the Keras (a high-level neural network API) framework, a short-term prediction model of the convergent deformation of the shield tube section at this location is established. The results show that the model successfully predicts the convergence of the tunnel section in the next 10 days, and the prediction accuracy reaches 93.6 %. The short-term prediction of key sections and the near warning sections is realized, so as to prevent it in advance.