基于支持向量回归的地铁深基坑

地表沉降预测

谭震霖

摘 要

近年来，随着我国城市轨道交通的大规模建设，地铁车站的建设得到快速发展，因此在地铁深基坑工程的施工过程中的安全问题不容忽视，而地铁深基坑开挖引起的周围地表沉降是实际地铁深基坑工程中的显著问题之一。其中，地铁深基坑开挖过程中的地表沉降监测与预测是地铁深基坑施工过程中最重要的工作之一。由于地铁深基坑的监测反馈具有滞后性，因此通过地铁深基坑地表沉降的预测来及时调整施工方案十分重要。在对于地铁深基坑地表沉降预测中，由于监测数据具有“样本数量少”、“预测精度要求高”等特点，而机器学习算法能较好处理地铁深基坑地表沉降监测数据的特点，本文通过机器学习算法、地铁深基坑理论分析与实际监测的累计沉降数据，将地铁深基坑与周围地表视为一个整体进行研究，在探究地铁深基坑地表沉降的影响因素基础上，选择基于多核SVR的地铁深基坑地表沉降预测。

首先，本文整理了国内外关于地铁深基坑地表沉降影响因素的相关理论，建立了地铁深基坑地表沉降评价指标体系，并对地铁深基坑地表沉降影响因素建立了量化标准表。然后，根据《地铁深基坑地表沉降影响因素相对重要性问卷调查》，通过对地铁工程师进行调查问卷、施工现场咨询相结合的方法，结合层次分析法计算，得出地铁深基坑地表沉降影响因素评价指标体系中的各个影响因素初始权重值。最后，选取武汉市某地铁深基坑，基于该工程概况，对每个影响因素相应的赋值，选取该基坑第八段土方开挖的四个监测点对应累计沉降监测数据。在通过线性支持向量回归来预测地铁深基坑地表沉降，并将地铁深基坑地表沉降各个影响因素的权重，作为初始化的线性支持向量回归模型来预测地表沉降，并与基于非线性支持向量回归下地铁深基坑地表沉降预测进行结果对比，结果显示多核的非线性支持向量回归对地铁深基坑地表沉降预测精准度更高。

本文根据地铁深基坑地表沉降的监测数据，建立了地铁深基坑地表沉降的SVR模型。在地铁深基坑施工时，有利于施工单位根据预测结果及时调整施工方案，以减少地铁深基坑土方开挖时的安全事故。该预测模型对地铁深基坑建设具有重要的工程应用意义，对今后类似工程具有借鉴价值。

关键词：地铁深基坑    地表沉降   线性SVR   非线性多核SVR  

Abstract

In recent years, with the large-scale construction of urban rail transit in China, the construction of subway station deep foundation pit is increasing day by day. Therefore, the safety problem in the construction of subway deep foundation pit engineering attracts more and more attention. The monitoring and predicting of ground surface subsidence during the excavation of deep foundation pit is one of the most fundamental element. Due to the hysteresis in the monitoring and feedback of deep foundation pit in subway, it is fairly important to adjust the construction plan by predicting the surface subsidence of deep foundation pit in subway in advance. In the process of prediction, the monitoring data has some characteristics like “inadequate sample” and “strict requirement of forecast accuracy”. Whereas, machine learning algorithms can better handle those data. Based on machine learning algorithms, subway deep foundation pit of theoretical analysis and actual monitoring cumulative settlement data, this paper regards the subway deep foundation pit as well as the surrounding surface of the earth as a whole. On the basis of the investigation of the factors, the ground settlement of the subway deep foundation pit, this paper intends to choose multi-core SVR model as the prediction model of ground settlement.

First of all, this paper reviews the related theories about the factors influencing the surface settlement of deep foundation pits both at home and abroad, and gives out the evaluation index system for the surface settlement of deep foundation pits in subway. A quantitative standard table for the factors affecting the surface settlement of deep foundation pits in subway is established, too. Then, according to The subway deep foundation pit surface subsidence influencing factors of relative importance questionnaire survey, the study of questionnaire about the construction of the subway engineers, the on-site consulting, as well as ahp, it is concluded that the subway deep foundation pit of surface subsidence various factors in the evaluation index system of influencing factors of initial weights. Finally, a subway deep foundation pit in Wuhan city was selected. Based on the general situation of the project, the corresponding values of each influencing factor were assigned. Four monitoring points in the eighth section of the excavation of the foundation pit were selected to correspond to the accumulated settlement monitoring data. Through linear support vector regression to predict the surface settlement, the subway deep foundation pit and the weight of each factors in the subway deep foundation pit surface subsidence, as the initialization of the linear support vector regression model to predict the surface subsidence, and based on the nonlinear support vector regression subway deep foundation pit under the surface subsidence prediction results comparison, the results showed that predicting precision of the nonlinear of multi-core support vector regression for subway deep foundation pit surface subsidence is higher.

Based on the monitoring data of surface subsidence of deep foundation pit in subway, the SVR model of surface subsidence of deep foundation pit in subway is established in this paper. In the construction of deep foundation pit, it is beneficial for the construction unit to timely adjust the construction plan according to the predicted results, so as to reduce the safety accidents in the excavation of deep foundation pit. This prediction model has great engineering application significance for the construction of deep foundation pit of subway and has reference value for similar projects in the future.

Keywords: Subway deep foundation pit Surface subsidence Linear support vector regression Network Nonlinear multi-kernel support vector regression