Time series prediction of tunnel boring machine (TBM) performance during excavation using causal explainable artificial intelligence (CX-AI)

Abstract

Since early warning is significant to ensure high-quality tunneling boring machine (TBM) construction, a real-time prediction method based on TBM data is proposed. To solve the “black box” problem of prediction by artificial intelligence (AI) methods, the causal explainable gated recurrent unit (CX-GRU) is developed to achieve real-time prediction for TBM parameters. The approach is implemented in a tunnel construction project in Singapore and the results indicate that CX-GRU performs well with the R square score are 0.9140 and 0.9184 in real-time prediction for thrust force and soil pressure. The causal discovery component can increase the computational efficiency of model training by 8.8% on average. According to the SHAP analysis of prediction results, the thrust force is more sensitive to the input TBM features, while the soil pressure is more sensitive to historical data. The CX-GRU is more reliable and efficient when applied to TBM projects than traditional methods.