Rock-Machine Interaction Model in TBM Tunnels Based on Optimized Deep Learning Algorithms

Since tunnel boring machines (TBMs) are gradually applied in deep and long tunnel buildings, they could encounter many complex and geological circumstances like TBM jamming, water bursting, and other adverse disasters. Therefore, rock condition perception is urgent for the tunnel construction schedule and safety. With the assistance of machine learning technology, numerous TBM data can be handled rapidly by using ML, and it provides the potential for real-time rock condition perception and real-time supporting proposals to avoid severe tunnel damage and loss of property. Due to the relatively poor ability of long short-term memory (LSTM) to solve continuous prediction issues, a convolutional neural network (CNN) structure is added in front of LSTM, and a sparrow search algorithm is adopted to tune with the original CNN-LSTM, so the final model mainly relies on SSA-CNN-LSTM. It can achieve the score (ACC=0.73368 in the train data, ACC=0.71957 in the validation data, and ACC=0.66667 in the test data). After model evaluation and comparisons with other models, the built model behaves the best and is followed by SSA-MLP, PSO-CNN-LSTM, SSA-SVM, SSA-LSTM, and LSTM in sequence.