Accurate Prediction of Tunnel Face Deformations in the Rock Tunnel Construction Process via High-Granularity Monitoring Data and Attention-Based Deep Learning Model

Abstract

Monitoring and predicting the deformation of surrounding rocks in the rock tunnel construction process is of great significance. This study implemented a wireless sensor network (WSN), including gateway transmission, relay point, and sensor nodes, to obtain high granularity deformation data during construction. A transformer model is proposed, which considers the construction sequence into the positional embedding and has an attention module to deeply learn the high dimensionality correlation between the nearby deformation data and the tunnel face deformation. The attention-enhanced LSTM model and the LSTM model are also constructed to compare them with the performance of the transformer model. A site study conducted on a shallow buried tunnel section suggested an excellent performance of the proposed WSN system. The transformer model shows the best performance in terms of the model prediction results, which can extract more information from the time sequence data than the attention-enhanced LSTM and LSTM models. The proposed system has great value as guidance and reference for the construction of rock tunnel projects in complex and unfavourable geological conditions.

Keywords

• rock tunnel construction
• tunnel face deformation
• tunnel monitoring
• deep learning model
• attention module