Artificial intelligence prediction of surface settlement induced by twin shields tunnelling
Gan Wang, Qian Fang, Jun Wang, Qiming Li, Haoran Song, Jinkun Huang
Abstract
The accurate prediction of the surface settlement trough is essential for the safety assessment of tunnel construction in densely occupied urban areas. In this study, we propose an artificial intelligence model to predict surface settlement troughs induced by twin tunnelling. The proposed model includes a newly proposed formula for describing settlement trough and a new calculation method of loss. The model uses a Graphical Convolutional Neural network (GCN) to extract latent feature information from field monitoring data that shows the state of the surrounding ground before the second shield passage. The proposed model is verified by comparing its results to those of two other models. The analysis shows that the developed calculation method of loss and consideration of the state of surrounding ground significantly improve the prediction accuracy of surface settlement troughs. While adding more monitoring points can offer benefits, the performance gains become weaker as the number of monitoring points increases. Therefore, we recommend using 24 monitoring points for the proposed model as it strikes the optimal balance between performance and computational efficiency.