Abstract
The relation between geological features of tunnel wall and groundwater discharge, that were observed during the excavation of the Seikan Under-Sea Tunnel was successfully analyzed by using the quantification and the artificial neural network (ANN) models. Data obtained from a part of 1.9km long of a pilot tunnel excavated from the Honshu side was used. This selected part was divided into small parts of 10m length, then the total amount of discharge, total width of fault zones, number of fractures, dominant geological features and so on observed in every small part were summarized. Then, 190 data sets presenting the combination of tunnel discharge and geological features for all small parts were prepared for the analysis. The analysis can be divided into two processes. One is the process determining best parameters in the quantification and the ANN models that can well reconstruct the discharge of the selected 30-95 small parts from geological data (Training phase). Another is the process estimating discharge expected in the other small parts that are not used in the training phase (Test phase).
It was found that the discharge can be well reconstructed by ANN model with the correlation coefficient of large than 0.9, when the discharge measured at the former small part was treated as a data as well as the geological features. The ANN model gave better result than the quantification model and seems to be more applicable. Although the general trend of discharge distribution was well predicated by using ANN model in the test phase, the correlation between the predicted and the observed discharges was not so high. Form this result, it can be concluded other geological data such as rock type, connectivity of fractures and so on, must be used for better prediction.
