Evolutionary hybrid neural network approach to predict shield tunneling-induced ground settlements

This study proposes an artificial intelligence approach to predict ground settlement during shield tunneling via considering the interactions among multi-factors, e.g., geological conditions, construction parameters, construction sequences, and grouting volume and timing. The artificial intelligence approach employs a hybrid neural network model that incorporates a differential evolution algorithm into the artificial neural network (ANN). The differential evolution algorithm is used to determine the optimized architecture and hyperparameters of ANN. The adaptive moment estimation (Adam) method is then employed to facilitate the training process of ANN. On the strength of Adam, the differential evolution algorithm is further enhanced to process a large number of ANN candidates without consuming massive computing resources. The proposed hybrid model is applied to a field case of ground settlements during shield tunneling in Guangzhou Metro Line No. 9. Geological conditions and shield operation parameters are first characterized and quantified by a feature extraction strategy, then input for the model. Results verifies the accuracy of prediction using the proposed hybrid model. Moreover, shield operation parameters with high influence on ground settlement are identified through a partial derivatives sensitivity analysis method, which can provide guidance for shield operation.