Neural networks as a means for predicting convergence in tunnels

Whereas a number of different analytical and numerical methods are presently available for calculating the size of rock deformations in the vicinity of a tunnel, the author’s aim has been to develop a new method for the prediction of maximum (H1) convergence, based on neural networks. Two different types of neural networks, a back-propagation network and an LVQ network, were investigated. Based on 11 selected input parameters, a correctly Iearnt neural network was in both cases developed, which was able, in the great majority of cases (more than 80 %), to reliably predict maximum convergence during tunnel construction. Comparisons between the results obtained using this method (for 4 tunnels, with 292 typical cross-sections) and those obtained using a suitably selected mathematical model also showed that these two methods can be used, together, for the selection and detailed design of support systems for particular tunnel sections.