Abstract:

Currently, shield operation monitoring relies heavily on the experience of operators. To overcome this limitation and boost the safety and stability of shield tunneling, a monitoring approach based on a double-layer long short-term memory network(LSTM) is proposed herein. Before modeling, data are preprocessed, and key feature variables, such as penetration and torque, along with their related variables, are selected from the operating data using the extreme gradient boosting algorithm and empirical knowledge. These variables then serve as inputs for the monitoring model, while the key feature variables of the shield serve as the outputs of this model. Next, a double-layer LSTM fault-monitoring model is developed using deep learning techniques to capture temporal correlation features within and between loops. Subsequently, monitoring statistics T² and E_spe are constructed, and different monitoring strategies are applied based on varying situations. Finally, the model is tested on the East Line project of the Yellow river tunnel to validate its effectiveness. Furthermore, the performance of this model is compared with that of an autoencoder model without temporal feature learning, a single-layer LSTM model, and other comparative algorithms. Experimental results reveal that the double-layer LSTM method achieves a false alarm rate of less than 1.25% and a 91.6% detection rate for a normal excavation ring and an abnormal excavation ring, respectively, thus validating its effectiveness in monitoring shield tunneling operations.

Key words: slurry shield, operation status monitoring, time-series analysis, double-layer long short-term memory network