Surface settlement modelling using neural network

Due to land shortage and rapid urbanization, most countries tap on underground resources to optimize land use. This allows valuable surface land to be used for livable uses and reduces the negative impact on city living. In Singapore, rail lines are constructed underground in soft soils at shallow depths. Tunnel Boring Machine (TBM) can be used for excavation of rail tunnels. However, TBM tunnelling activities may lead to surface settlement due to ground loss during tunnelling. To prevent surface settlement, forecasting methods are better than reactive measures. However, historical methods were limited in understanding the complex shield-ground interaction. As Artificial Intelligence gained popularity, the Artificial Neural Network (ANN) is a promising new forecasting method. However, ANN models would only be effective when key parameters are identified and used. In this study, literature research was carried out on these key parameters. After the key parameters were determined, a dataset extracted from three local tunnelling projects was used for analysis. The key parameters in the dataset were studied on their influence on uncertainty of settlement prediction. A simple sensitivity analysis which involved adding Gaussian Noise to the input patterns was conducted. Subsequently, ANN models with different input combinations and different number of hidden neurons were analyzed using an ANN model based on MATLAB. This is to determine an optimal ANN architecture with reasonably high model accuracy. This optimal model could be used to predict surface settlement for new tunnels under similar geological environments.Bachelor of Engineering (Civil