ABSTRACT
Using Machine Learning (ML) and Deep Learning (DL) algorithms to predict ground surface settlement caused by shield tunneling has being a hot topic. However, previous developed models are general weakness in the generalization performance when solving the problem in multi-regions. Because the data were usually collected based on a certain project, resulting in the limitations of the data in quantity and representativity. In this study, numerous data were collected from shield tunnels derived in various stratum conditions from different cities of China. Then, the improved models were proposed to identify the relationship between shield tunneling and the settlement. The efficiency of a DL algorithm Bi-directional Long Short-Term Memory (Bi-LSTM), ML algorithms Back-propagation Neural Network (BPNN) and Random Forest (RF) were compared. The results indicated that the accuracy of the proposed models have superior performance on the prediction accuracy of surface settlement, compared with dedicated model trained using individual city data. Furthermore, Bi-LSTM is demonstrated as a better algorithm to predict surface settlement than BPNN and RF.
