ABSTRACT
The slurry circulation system in slurry shield tunneling is crucial for maintaining tunnel face stability and cutting efficiency. In long-distance tunnel excavations, simultaneous operation of multiple pumps leads to parameter coupling, increasing control complexity. Imbalanced pipeline pressure risks efficiency and safety. Traditional manual control struggles with excessive information and adjustment difficulty, potentially triggering engineering risks. To address these challenges, this study employs the Peter and Clark Momentary Conditional Independence (PCMCI) algorithm for causal analysis of high-dimensional time-series tunneling data, clarifying causal relationships among key parameters in the slurry shield system. A Graph Convolutional Network (GCN) model is developed to capture complex network structures, enabling real-time and accurate prediction of the multivariable-coupled slurry system’s responses. This predictive model is then used to optimize system control parameters, ensuring safe and efficient management. The proposed method has been successfully applied in a metro tunnel project in Shanghai, yielding favorable results.
