Prediction for the future mechanical behavior of underwater shield tunnel fusing deep learning algorithm on SHM data

Abstract

Predicting the future mechanical behavior of tunnel structure is vitally important to prevent accident disasters. However, in most of the existing models, the inadequate consideration for influencing factors reduced the final prediction accuracy. To this end, this study aims to develop an accurate prediction model considering the coupling effects of multiple influencing factors. First, the framework of model integrates the effects of Temporal, Spatial, and Load (TSL) dependencies is developed based on deep learning algorithm. Subsequently, TSL is formulated on the monitoring data obtained from the Wuhan Yangtze River tunnel and used to predict the mechanical behavior of this study case under an extreme condition. Through a series of experiments, the necessity of considering the coupling effects of multiple influencing factors is verified, and the parameter effects on model predictive capability are discussed. In addition, some commonly used prediction models, such as RNN, LSTM, Xgboost, SVR, LR, are selected as baselines to compare with TSL. Experimental results indicate that the predictive ability of TSL is superior among all models, whose accuracy improves 2.853% in next 15 days prediction. Therefore, it is essential to consider the couple effects of multiple factors, and the presented model is reasonable.