Intelligent Approach to Estimation of Tunnel-Induced Ground Settlement Using Wavelet Packet and Support Vector Machines

This paper proposes a novel hybrid approach that integrates wavelet packet transformation (WPT) and least-squares support vector machines (LSSVMs) to enhance the accuracy and reliability regarding the estimation of tunnel-induced ground settlement on a daily basis. The original time-domain signal, measured settlements over a given time period, is decomposed into a series of sequences using WPT. LSSVM models are then built to predict the target sequences within high- and low-frequency regions. The predicted sequences are reconstructed to recover the estimated tunnel-induced ground settlement over time. Two indicators, mean absolute error (MAE) and root mean square error (RMSE), are proposed to illustrate the correspondence between individual pairs of model predictions and actual observations for performance analysis. A realistic tunnel case in the Wuhan, China, metro system is utilized to demonstrate the feasibility and applicability of the proposed WPT-LSSVM approach. Comparisons between existing methods and the developed approach are analyzed and discussed in detail. Results indicate that SVMs display higher prediction accuracy than artificial neural networks in estimating tunnel-induced ground settlement, and the proposed WPT-LSSVM approach has higher accuracy and reliability than the traditional LSSVM approach. This approach can be implemented as a decision-making tool for the time-series analysis and estimation of tunnel-induced settlement, which can provide support for improving safety assurance in tunneling projects. (C) 2016 American Society of Civil Engineers.