Modern Machine Learning Techniques for Univariate Tunnel Settlement Forecasting: A Comparative Study

Abstract

Tunnel settlement commonly occurs during the tunnel construction processes in large cities. Existing forecasting methods for tunnel settlements include model-based approaches and artificial intelligence (AI) enhanced approaches. Compared with traditional forecasting methods, artificial neural networks can be easily implemented, with high performance efficiency and forecasting accuracy. In this study, an extended machine learning framework is proposed combining particle swarm optimization (PSO) with support vector regression (SVR), back-propagation neural network (BPNN), and extreme learning machine (ELM) to forecast the surface settlement for tunnel construction in two large cities of China P.R. Based on real-world data verification, the PSO-SVR method shows the highest forecasting accuracy among the three proposed forecasting algorithms.

Tunnel Surface Settlement Forecasting with Ensemble Learning

“…Different from traditional forecasting problems, tunnel surface settlement forecasting has the challenges/properties of a short-period of time data available, univariate training data, and various hidden factors that are missing in the dataset. Existing works, such as [5], have shown the instability of prediction models while only single type models are used. To improve the robustness of the final prediction model, an integration of multiple models using ensemble learning algorithms is desired.…”

“…In this study, the original Adaboost.RT algorithm was altered by selecting different types of base learners to obtain the final ensemble generalized learning framework. According to our previous study on this topic [5], three base classifiers, namely, BPNN, ELM, and SVR were selected to build the ensemble learner.…”

“…In 2019, Hu et al [5] compared most existing ML techniques for ground surface settlement prediction in real-world tunnel construction processes and pointed out that deep learning techniques, such as convolutional neural networks (CNNs) and long short-term memory (LSTM) neural networks, are not suitable for analyzing short sequence time series data, such as tunnel settlement data [6]. This study is a follow-up work of [5]. There are two main difficulties for tunnel settlement forecasting.…”

“…Machine learning (ML), representing one of the hottest topics in data-driven methods, has achieved great successes in recent research studies. Compared to model-based methods, which require expert knowledge and experience to build physical or mathematical models, ML methods predict the future tunnel surface settlement purely based on historical data [5]. The established ML models are usually much more complex than the conventional mathematical model and can hardly be interpreted using several mathematical equations.…”

“…In contrast, traditional ML techniques are more reliable prediction methods for tunnel settlement in the short-term. In a study by Hu et al [5], through a series of comparative study, three machine learning techniques, namely, back-propagation neural network (BPNN) [8], extreme learning machine (ELM) [9], and support vector regression (SVR) [10], were selected as best predicting the tunnel settlement in the short-term for various cases. Following this study [5] and using the same dataset, in this study, we show that there exists a complex nonlinear relationship between tunnel settlement and many random uncertain factors; it is difficult to predict the tunnel settlements using one single machine learning technique.…”