盾构掘进地表沉降量化分析成套方法与系统设计研究

孙澜

摘 要

随着盾构法在隧道建设中的广泛应用，工程中对盾构施工安全性的要求也越来越高，国内许多城市开始研发地铁施工安全相关的数据平台来对盾构施工过程进行更精细化的安全管理。盾构掘进过程中诱导的沉降是施工过程中常见的安全问题之一，精准预测地表沉降是隧道工程中的重要问题之一。针对该问题，国内外众多学者提出了多种预测方法，但还未形成良好的研究体系。本文研究了三种沉降预测方法，并分析其特点及适用条件，将相关成果嵌入到地铁平台中的地表沉降预测模块。本研究以某市某盾构施工地铁区间为工程背景，主要进行了以下工作：

（1）采用Peck公式及其修正方法法和Plaxis 3D 数值计算法预测地表沉降，建立了针对该研究隧道区段的预测模型。结果表明，Peck公式法计算简便，结果合理，但存在较强的地域性，需要进行修正。Plaxis 3D数值计算法仿真性能强大，预测结果准确度高且可靠性强，但建模难度较高、计算耗时。

（2）建立基于梯度提升决策树（Gradient Boosting Decision Tree，GBDT）的沉降预测模型，通过嵌入法-GBDT模型得到了与沉降相关的64个影响因素的重要性排列，研究发现选取相关性最高的9个特征时，模型表现最优。通过调整损失函数、网格搜索方法优化超参数，进一步地提升了模型预测性能。研究结果表明，人工智能方法能够有效地解决岩土工程面临的非线性问题，预测精确度高、速度快，但建模过程没有很好的考虑土体力学响应，对于工程中的问题解释性不强。

（3）基于对三种方法准确度、建模耗时、计算速度、参数数目、可视化程度、普适性、可靠性和影响因素个数等各维度的评估，提出将安全风险等级映射到沉降预测方法的策略，使得本研究满足不同沉降安全风险等级下对预测方法的需从定性分析到定量分析，归纳各预测方法在不同工况下的使用建议，并辅以实例验证。

本研究基于某市某隧道工程项目，分析对比了经验公式法、数值计算法和人工智能方法在工程中的应用，总结各方法的适用工况并构建了通用高效的沉降预测工具框架，并采用多维度对提出的方法进行评估，研究成果具有实际的工程意义和较高的应用价值，对于隧道施工信息化建设有参考价值。

关键词： 沉降预测；机器学习；沉降公式；数值计算

Abstract

With the wide application of the shield method in tunnel construction, the requirements for shield construction safety are getting higher and higher. Many cities in China have begun to develop data platforms related to subway construction safety to make the shield construction process more precise. security management. The settlement induced in the process of shield tunneling is one of the common diseases in the construction process. Accurate prediction of surface settlement is one of the important problems in tunnel engineering. In response to this problem, many scholars at home and abroad have proposed a variety of prediction methods, but a good research system has not yet been formed. This paper studies three subsidence prediction methods, analyzes their characteristics and applicable conditions, and embeds the relevant results into the surface subsidence prediction module of the subway platform. This study takes a shield tunnel construction subway section in a city as the engineering background, and mainly carries out the following work:

(1) The Peck formula and its correction method and the Plaxis 3D numerical calculation method are used to predict the surface subsidence, and a prediction model for the study tunnel section is established. The results show that the Peck formula method is easy to calculate and the results are reasonable, but there is a strong regionality, which needs to be revised. Plaxis 3D numerical calculation method has powerful simulation performance, high accuracy and reliability of prediction results, but high modeling difficulty and time-consuming calculation.

(2) Establish a settlement prediction model based on Gradient Boosting Decision Tree (GBDT), and obtain the importance ranking of 69 influencing factors related to settlement through the embedding method-GBDT model. With 9 features, the model performs best. The model prediction performance is further improved by adjusting the loss function and optimizing the hyperparameters using grid search methods. The research results show that the artificial intelligence method can effectively solve the nonlinear problems faced by geotechnical engineering, with high prediction accuracy and fast speed. powerful.

(3) Based on the evaluation of the accuracy of the three methods, modeling time, calculation speed, number of parameters, degree of visualization, universality, reliability, and number of influencing factors, it is proposed to map the safety risk level to settlement. The strategy of the prediction method makes this study meet the needs of the prediction method under different subsidence safety risk levels. From qualitative analysis to quantitative analysis, the suggestions for the use of each prediction method under different working conditions are summarized and verified by examples.

Based on a tunnel engineering project in Wuhan, this research analyzes and compares the application of empirical formula method, numerical calculation method and artificial intelligence method in engineering, summarizes the applicable working conditions of each method, and builds a universal and efficient settlement prediction tool framework. The proposed method is evaluated in terms of dimensions. The research results have practical engineering significance and high application value, and have reference value for the construction of tunnel construction informatization.

Key words: Settlement prediction; machine learning; settlement formula; numerical simulation