Wavenet ability assessment in comparison to ANN for predicting the maximum surface settlement caused by tunneling

Abstract

An alternative method of maximum ground surface settlement prediction, which is based on integration between wavelet theory and Artificial Neural Network (ANN), or wavelet network (wavenet), is presented. In order to minimize the risk of tunneling, a tunnel engineer needs to be able to make reliable prediction of ground deformations induced by tunneling. Any prediction from numerical analysis was highly dependent on the model adopted for modeling the soil behavior. However, setting up a realistic model that would be able to calculate tunneling-induced settlement profiles is rather difficult. Most of the researches show that the capability (i.e. pattern recognition and memorization) of an ANN is suitable for inherent uncertainties and imperfections found in geotechnical engineering problems considering its successful application without any restriction. Wavenet is a single hidden layer feedforward neural network, which uses wavelets as its activation functions. In this study different wavelets are applied as activation functions to predict the maximum surface settlement due to tunneling. Wavenet parameters such as dilation and translation are fixed and only the weights of the network are optimized during its learning process, which is performed by a back-propagation algorithm. The efficacy of this type of network in function learning and estimation is demonstrated through measurements extracted from EPB shield tunneling. The simulation results indicate decrease in estimation error values that depicts its ability to enhance the function approximation capability and consequently exhibits excellent learning ability compared to the conventional back-propagation neural network with sigmoid or other activation functions.