Data-driven real-time prediction for attitude and position of super-large diameter shield using a hybrid deep learning approach

• DOI: 10.1016/j.undsp.2023.08.014
• Publication Date: 2023-10-17T01:26:45Z
• Authors: Yanbin Fu; Lei Chen; Hao Xiong; Xiangsheng Chen; Andian Lu; Yi Zeng; Beiling Wang
• Keywords: Position (finance)

The presented research introduces a novel hybrid deep learning approach for the dynamic prediction of attitude and position super-large diameter shields - critical consideration construction safety tunnel lining quality. This study proposes predicting shield. consists principal component analysis (PCA) temporal convolutional network (TCN). former is used employing feature level fusion based on features shield data to reduce uncertainty, improve accuracy effect, 9 sets required characteristic are obtained. latter adopted process sequence in advantages potential convolution network. approach's effectiveness exemplified using from project China. obtained results show remarkable global position, with an average error ratio less than 2 mm four outputs 97.30% cases. Moreover, displays strong performance accurately sudden fluctuations 89.68%. proposed model demonstrates superiority over TCN, long short-term memory (LSTM), recurrent neural (RNN) multiple indexes. Shapley additive exPlanations (SHAP) also performed investigate significance different process. provides real-time warning driver adjust shields.