Hard-rock tunnel lithology prediction with TBM construction big data using a global-attention-mechanism-based LSTM network

Abstract

The TBM-constructed rock tunnel often suffers from low comparability of efficiency between geological condition detection and the TBM real-time operation requirements. This article developed a time-related intelligent model for tunnel lithology prediction using TBM construction big data with illustration by the Yinsong Water Diversion Project. The global attention mechanism based long-short-term-memory (LSTM) network was established to model the cyclic TBM construction data and make predictions of lithology at the tunnel face with 9 selected featuring input parameters. The established global-attention-mechanism-based LSTM network was found to predict well tunnel lithology with TBM construction data and outperformed the conventional LSTM network and other models in accuracy and F1-scores. The results could help the TBM drivers to adjust the operational parameters at real time for high-efficient tunnel construction. With implementation of the establish model into the TBM operational system, we will develop intelligence construction mode for the long TBM-constructed hard rock tunnels.