Modeling of Tunneling Total Loads Based on Symbolic Regression Algorithm

Abstract

The tunneling total load is one of the core control parameters for safe and efficient construction using tunneling machines. However, because the tunneling process involves complex coupling relationships between the equipment and the local geology, theoretical derivation is difficult. The development of tunneling data detection and acquisition technology has led to extensive load modeling based on data analysis and machine learning. However, it is difficult to obtain an explicit interpretable model that satisfies certain physical rules. In this paper, a modeling method based on symbolic regression is proposed. The method mainly includes three modules: construction of π quantities, feature selection, and model training. Through dimensional analysis, the π quantities are constructed so as to impose physical constraints on the training process. Feature selection based on a nonlinear random forest model is used to improve the modeling efficiency. Finally, an explicit nonlinear load model is obtained using symbolic regression, which satisfies the basic equilibrium theory of mechanics and the dimensional rules of physics. The proposed approach is compared with general linear regression and an artificial neural network. The results show that the proposed method produces a load model that is interpretable and accurate, providing an excellent reference for construction excavation.