A coupled deep learning approach for shield moving performance prediction of underground tunnel construction

Abstract

Recently, deep learning methods have attracted great attentions for geotechnical engineering, especially for the safety construction of underground tunnels. In this regard, the control and regulation of operational parameters are vital for safety and intelligent tunnel construction. In this study, hybrid data-driven models on the basic of optimization approaches [particle swarm optimization (PSO), multilayer hierarchical heterogeneous PSO] and deep learning methods (long short-term memory and gated recurrent unit neural network) were developed. The optimization approaches were utilized to obtain the optimal parameters of deep learning methods. Apart from that, geological data sets were extended to match the scale of operational data sets using Kriging method, which considered the effects of local geology comprehensively. The developed hybrid models (DHMs) were utilized to estimate the speed of earth pressure balance (EPB) shield machine for a tunnel project in China. Results indicate that errors in estimated results by DHMs are acceptable in geotechnical engineering. Besides, the DHMs exist potentials to cope with time-series data sets from underground tunnel project. After that, Pearson correlation coefficient was used to evaluate the relative importance of influential factors on target variable. Results show that penetration has significant correlation relationship with excavation speed. The geological and operational parameters are equally important for the prediction performance of DHMs. The developed DHMs can provide a reference for regulating shield moving performance and attaining safety construction of underground tunnels.