ICA-ANN, ANN and multiple regression models for prediction of surface settlement caused by tunneling

AuthorsMohammad Reza Moghaddasi, Majid Noorian-Bidgoli
JournalTunnelling and Underground Space Journal
Paper TypeFull Paper
Published At2018
Journal GradeScientific - research
Journal TypeTypographic
Journal CountryIran, Islamic Republic Of
Journal IndexISI ,SCOPUS ,ISC

Abstract

Nowadays, with increasing urbanization and population of cities, the amount of internal transportations enlarged. To facilitate these movements, need for subway tunnels has been considerably increased. In urban areas, subway tunnels are excavated in shallow depth under thick populated areas and soft ground. Its associated hazards include poor ground condition, presence of water table above the tunnel, shallow overburden and surface settlement induced by tunneling. To avoid damage to surface structures and environmental problems, Maximum surface settlement (MSS) and its accurate prediction is one of the serious challenges during this procedure. In this paper, a new hybrid model of artificial neural network (ANN) optimized by Imperialist competitive algorithm (ICA), called ICA-ANN, has been presented for the prediction of MSS. For this purpose, a total number of 143 datasets including, horizontal to vertical stress ratio, cohesion and Young’s modulus considered as input parameters and their corresponding MSS considered as an output parameter, were inquired from the line No. 2 of Karaj subway, in Iran. This datasets used in order to construct the MSS predictive models. To show the capability of the ICA-ANN model in predicting MSS, an ANN model and traditional statistical model of multiple regression (MR) was also employed. In order to assess the prediction performance of mentioned models, performance indices including, correlation coefficient (R2), root mean square error (RMSE) and variance account for (VAF) were calculated. Results of comparing reveals that the proposed ICA-ANN model is capable to predict MSS with higher reliability than the ANN and MR models.