Surface settlement modelling using neural networks

Singapore is a highly urbanized and densely populated city with high-rise buildings and complex infrastructure above and below the ground. Hence, it is imperative that the ground movements induced from tunnelling operations are anticipated and controlled prior to their occurrence, which might otherw...

Full description

Saved in:
Bibliographic Details
Main Author: Kori, Prajwal Jagadeesh
Other Authors: Zhao Zhiye
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2022
Subjects:
Online Access:https://hdl.handle.net/10356/163847
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Singapore is a highly urbanized and densely populated city with high-rise buildings and complex infrastructure above and below the ground. Hence, it is imperative that the ground movements induced from tunnelling operations are anticipated and controlled prior to their occurrence, which might otherwise produce undesirable consequences that can be catastrophic in nature and this may even extend to human lives. This alludes to critical planning of tunnelling activities that should be accounted for into the overall construction process to prevent loss of lives and property destruction. Previously, engineers and research scientists relied on empirical methods and analytical methods to predict ground settlement, but in essence they have their own limitations and do not accurately predict the settlement values, especially given the growing complexity of the underlying transportation infrastructure, adjacent building foundations that have varying ages and underground utilities. This study investigates the use of artificial neural networks (ANN) and explores different dimensionality reduction techniques to achieve improvements in predicting settlement values. The dataset was collected across three tunnelling projects in Singapore and encompasses a plethora of geological and TBM operational parameters. The datasets were fed to ANN models with different hyperparameter configurations. The performance measures of each model fed with different dataset obtained from different dimensionality reduction techniques were evaluated and the model yielding the best results was chosen.