3D prediction of tunneling-induced ground movements based on a hybrid ANN and empirical methods

• Main Authors: Hajihassani, M., Kalatehjari, R., Marto, A., Mohamad, H., Khosrotash, M.
• Format: Article
• Published: 2019
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85061482262&doi=10.1007%2fs00366-018-00699-5&partnerID=40&md5=6e0068c682d37379a967e95f9d65485d; http://eprints.utp.edu.my/22283/

Tunnel construction in urban areas causes ground displacement which may distort and damage overlying buildings and municipal utilities. It is therefore extremely important to predict tunneling-induced ground movements in tunneling projects. To predict the tunneling-induced ground movements, artificial neural networks (ANNs) have been used as flexible non-linear approximation functions. These methods, however, have significant limitations that decrease their accuracy and applicability. To overcome these problems, the use of optimization algorithms to train ANNs is of advantage. In this paper, a hybrid particle swarm optimization (PSO) algorithm-based ANN is developed to predict the maximum surface settlement and inflection points in transverse and longitudinal directions. Subsequently, the transverse and longitudinal troughs were obtained by means of empirical equations and 3D surface settlement troughs were ploted. For this purpose, extensive data consisting of measured settlements from 123 settlement markers, geotechnical properties and tunneling parameters were collected from the Karaj Urban Railway Project in Iran. The optimum values of PSO parameters were determined with the help of sensitivity analysis. On the other hand, to find the optimal architecture of the network, trial-and-error method was used. The final hybrid model including eight inputs, a hidden layer and three outputs was used to predict transverse and longitudinal tunneling-induced ground movements. The results demonstrated that the proposed model can very accurately predict three-dimensional ground movements induced by tunneling. © 2019, Springer-Verlag London Ltd., part of Springer Nature.