Real-Time Prediction of Strata Conditions in Front of EPB Shield based on Bayesian Algorithms

Abstract Prediction of strata conditions is one of the essential tasks in shield tunneling. the variability of strata and the uncertainty of construction bring greater challenges to prediction. The uncertainty was introduced to improve the accuracy and reliability of results. By combining the Bayesian algorithms with machine learning method, the prediction of strata conditions in front of shield machine was performed in this paper, based on the shield parameters. According to the engineering operation logic and correlation analysis, six main shield parameters are selected as input parameters. Thereafter, the Bayesian SoftMax Regression and Bayesian Neural Network with Markov Monte Carlo and Variational inference are adopted for the prediction. Accuracy index (Acc) and uncertainty validity index (Acc-prob) are proposed to evaluate the performance of Bayesian models. As the conclusion, Bayesian algorithms can effectively improve the prediction accuracy and provide reliable guidance for prediction of strata conditions in EPB shield. Simultaneously, it is crucial to select appropriate Bayesian models and inference methods tailored to the dataset's scale.