Neural-network-based regression model of ground surface settlement induced by deep excavation

Ground surface settlement is an important field measurement in deep excavation. The monitoring data are adopted to evaluate construction performance and to avoid large surface settlements incurred to adjacent structures. Due to the complicated geotechnical and construction factors affecting ground surface settlement, no single analytical method can accurately forecast ground surface settlement induced by deep excavation. This paper presents an artificial-neural-network-based (ANN-based) regression approach to the prediction of ground surface settlement induced by deep excavation. Case data of deep excavation projects recently finished in Taiwan were used to establish the model. Soil and construction-related parameters having significant influences on surface settlement were filtered to train and test the ANN. Validation was also performed to show that the ANN outperformed the multiple linear regression method in predicting ground surface settlement. The ANN-based forecast model can reasonably predict the magnitude, as well as the location, of maximum ground surface settlement induced by deep excavation.