ABSTRACT
Rokunohe tunnel of Tohuku Shinkansen project in Japan. The geological condition and the measurement point of the tunnel are illustrated in Figure 5. (a). Finite element meshes are shown in Figure 5. (b). Simulation has been performed in several computational steps for excavation of the tunnel top heading. FEM output selected by filed measurement point during training. Figure 6 illustrates the ANN structure for design parameter identification, (a), the ANN structure for subsidence prediction, (b). The sigmoid function is preferred as the transfer function in this study. The validation of a trained ANN model is based upon one or more error indices. In this paper, training parameter is used as a tolerance to indicate end of training which was defined when the number of training (Epoch) goes over 400000 or the error of the network (SSE) becomes below 0.0001.
