• CSCD核心中文核心科技核心
  • RCCSE(A+)公路运输高质量期刊T1
  • Ei CompendexScopusWJCI
  • EBSCOPж(AJ)JST
二维码

隧道建设（中英文） ›› 2026, Vol. 46 ›› Issue (3): 599-608.DOI: 10.3973/j.issn.2096-4498.2026.03.013

• 研究与探索 • 上一篇    下一篇

基于掘进参数的砂土地层盾构渣土改良效果分级与预测

郭栩含， 赵文*， 柏谦， 李佳龙   

  1. （东北大学资源与土木工程学院， 辽宁 沈阳 110819）
  • 出版日期:2026-03-20 发布日期:2026-03-20
  • 作者简介:郭栩含（2000—），女，辽宁朝阳人，东北大学土木工程专业在读硕士，研究方向为智能化土压平衡盾构渣土改良。E-mail：guoxuhan2000@163.com。 *通信作者： 赵文， E-mail：zhaowen@mail.neu.edu.cn。

Classification and Prediction of Soil Conditioning Effects for Shield Tunneling in Sandy Strata Based on Tunneling Parameters

GUO Xuhan, ZHAO Wen*, BAI Qian, LI Jialong   

  1. (School of Resources & Civil Engineering, Northeastern University, Shenyang 110819, Liaoning, China)
  • Online:2026-03-20 Published:2026-03-20

PDF

46

摘要： 为解决传统土压平衡盾构渣土改良效果评价方法依赖室内试验结果、指标分析相互独立，以及现有掘进参数预测模型未考虑改良剂参数的局限，提出一种同时考虑掘进参数、地层参数与改良剂参数的渣土改良效果分级与预测模型。依托沈阳地铁3号线砂土地层盾构区间1 372环掘进数据，选取场切深指数（IFP）、转矩切深指数（ITP）和贯入度（S）作为渣土改良效果评价指标，经数据清洗、异常值剔除及标准化处理后，通过K-means聚类算法建立“优、良、中、差”4个等级的渣土改良效果评价模型。在此基础上，进一步融合地层参数与改良剂参数，构建随机森林（RF）、极端随机树（ET）和梯度提升回归（GBR）3种模型对掘进参数进行预测。结果表明： 1）K-means聚类算法能有效区分改良效果等级； 2）ET模型对掘进参数的预测性能最优，对IFP、ITP、S预测的决定系数R2分别为0.877 1、0.881 2、0.843 4； 3）基于测试集验证的分级-预测模型总体准确率为80.69%。

关键词: 土压平衡盾构, 渣土改良, 掘进参数, 地层参数, 改良剂参数, K-means聚类算法, 极端随机树

Abstract:

Traditional methods for evaluating soil conditioning effects in earth pressure balance (EPB) shield tunneling rely heavily on laboratory tests and independent index analysis. However, existing tunneling parameter prediction models often neglect conditioner parameters. To address these limitations, the study proposes a classification and prediction model for soil conditioning effects that integrates tunneling parameters with stratum and conditioner parameters. The data is obtained from 1 372 rings in the sandy strata of the EPB shield-bored section of the Shenyang metro line 3. Based on the acquired data, the following parameters are selected as evaluation indicators for soil conditioning effects: the field penetration index IFP, torque penetration index ITP, and penetration degree (S). Following data cleaning, outlier removal, and standardization, a four-level evaluation model (excellent, good, medium, and poor) is established using the K-means clustering algorithm. Additionally, random forest, extremely randomized trees (ET), and gradient boosting regression models are built by combining stratum and conditioner parameters to predict tunneling parameters. The results show that (1) K-means clustering effectively differentiates between levels of conditioning effects; (2) the ET model outperforms others in predicting tunneling parameters, achieving coefficient of determination (R2) values of 0.877 1, 0.881 2, and 0.843 4 for IFP, ITP, and S predictions, respectively; and (3) the overall accuracy of the classification-prediction model, validated on the test set, is 80.69%.


Key words: earth pressure balance shield, soil conditioning, tunneling parameters, ground parameters, conditioner parameters, K-means clustering, extremely randomized trees

主办单位：中铁隧道勘察设计研究院有限公司    主管单位：中铁隧道局集团有限公司

本系统由北京玛格泰克科技发展有限公司设计开发 技术支持：support@magtech.com.cn

粤ICP备19038941号