A Long-term Probabilistic Forecasting Approach of TBM Operating Parameters based on Deep Learning
The work was presented in the the 4th International Conference on Information Technology in Geo-Engineering (4ICITG). The presentation video can be found here and the slides can be found here.
Abstraction: In tunnel construction, tremendous data of Tunnel Boring Machine (TBM) operating parameters will be produced, which makes the automatic construction based on data-driven models possible. The work develops a recurrent neural network (RNN) -based pipeline for probabilistic forecasting of the trend of TBM operating parameters in the next one minute in real-time from historical 40-second tunneling data. In the Jilin Yinsong Water Tunnel dataset, the cutter head torque and thrust prediction accuracies are more than 82% and 93%, respectively. The model shows strong potential to give long-term and real-time guidance for TBM drivers in practice and thus reduces the uncertainties in the TBM excavation process.
Keywords: Tunnel Boring Machine, Probabilistic Forecasting, Deep Learning
The project was implemented with python 3.9
pip install -r requirements.txt