Shield Tunnel (Segment) Uplift Prediction and Control Based on Interpretable Machine Learning

Abstract

Shield tunnel segment uplift is a common phenomenon in construction. Excessive and unstable uplift will affect tunnel quality and safety seriously, shorten the tunnel life, and is not conducive to the sustainable management of the tunnel’s entire life cycle. However, segment uplift is affected by many factors, and it is challenging to predict the uplift amount and determine its cause accurately. Existing research mainly focuses on analyzing uplift factors and the uplift trend features for specific projects, which is difficult to apply to actual projects directly. This paper sorts out the influencing factors of segment uplift and designs a spatial-temporal data fusion mechanism for prediction. On this basis, we extract the key influencing factors of segment uplift, construct a prediction model of segment uplift amount based on Extreme Gradient Boosting (XGBoost) v2.0.3, and use SHapley Additive exPlanation (SHAP) v0.44.0 to locate factors affecting uplift, forming an Auxiliary Decision-making System for Segment Uplift Control (ADS-SUC). An ADS-SUC not only detects the sudden change of the segment uplift successfully and predicts the segment uplift in practical engineering accurately, it also provides a feasible method to control the uplift in time, which is of great significance for reducing the construction risk of the tunnel project and ensuring the quality of the completed tunnel.