A novel real-time torque prediction of epb shield in mixed ground using machine learning method based on geological knowledge fusion

Intelligent tunneling has become a necessary technology in urban underground development. Machine learning (ML) algorithms have been widely used in predicting earth pressure balance shield (EPBS) machine tunneling; however, there is still a problem of the insufficient generalization ability of the prediction model so far. The complex strata lead to the shield–soil system becoming intricate and bring challenges for real-time prediction. Therefore, this paper proposes a prediction model based on geological knowledge fusion to solve the generalization problem. The soil mechanism (i.e., strength theory) is introduced to ML algorithms for the first time. Statistical analysis on shield operating parameters is carried out, and the geological survey is sorted out before training. Then, the input geological parameters generated by soil mechanics theories and operating parameters are trained by a long short-term memory (LSTM) neural network. The results showed that the model with geological knowledge fusion performs better than the model with only shield operating parameters in the complex strata. It was also found that using existing geotechnical knowledge and geology surveys can significantly improve the prediction ability of the model when the EPBS enters unfamiliar complex strata. The research method is promising and could be applied to the other prediction issues in complex boundary conditions of geotechnical engineering.