A parameter design model for EPB shield soil conditioning in gravel stratum based on machine learning

Abstract

The utilization of soil conditioners plays a vital role in preserving excavation safety during the construction of an earth pressure balanced (EPB) shield, as they enhance the reliability of the cutting chamber. The kinds and amounts of soil conditioner that ought to be pumped into the EPB system, however, lack theoretical direction. This study aims to develop predictive and design models to investigate the reciprocal relationship between soil conditioners and soil characteristics. The soil conditioner parameters were designed with the targeted characteristics of the soil using the intelligent hybrid design model, which combined the prediction model and the design model. The study's conclusions showed that the Gradient Boosting Regression (GBR) model, which was developed, had improved generalization and accuracy abilities when used to forecast the properties of upgraded soils. The current Artificial Neural Network (ANN) design model provided a variety of available options, but its accuracy was poor, and it could not be used directly. A hybrid GBR-ANN model created by the researchers showed a difference between the goal value and the experimental value acquired from laboratory soil testing of less than 10%. A solution that was required was the model's time-consuming explosion of the marginal value design. As a result, it is determined that using the Hybrid GBR-ANN model is adequate for figuring out the necessary amount and composition of soil conditioner, making it a workable method for soil property-oriented design.