基于掘进参量反演的TBM围岩等级预测识别方法研究

Prediction and Identification Method of Tunnel Boring Machine Surrounding Rock Grade Based on Tunneling Parameters Inversion

为探究TBM掘进参量和围岩等级关系,达到岩机信息互馈感知动态调整,辅助TBM主司机优化调整掘进参数的目的。采用数据清洗、分布统计和智能预测等手段,建立了一种基于自组织神经网络聚类和最小二乘支持向量机相结合的围岩等级预测识别方法(SOM-SVM)。主要结论如下:1)单个完整TBM掘进循环可分为空推段、上升段与稳定段,各掘进参量近似服从正态分布关系。2)推力切深指数(FPI)和刀盘转矩旋转切深指数(TPI)可反映隧道岩石掘进难易程度,FPT、TPI和围岩等级近似呈线性关系,可用该参量作为岩机敏感因子反演预测识别围岩等级。3)干扰异常数据样本点的预处理对SOM-SVM围岩预测模型收敛中心和波动半径有一定影响,数据预处理是保证围岩等级预测识别准确的关键。4)经标准试验数据样本和工程数据验证,不同的支持向量机核函数对围岩等级预测识别影响很大,线性核、多项式核、高斯径向基核函数围岩综合识别率分别为70.8%、81.2%、87.6%,围岩等级预测识别模型预测精度高、鲁棒性好。

The relationship between tunnel boring machine(TBM) tunneling parameters and surrounding rock grades is necessary to attain dynamic adjustment of rock machine information mutual feedback perception as well as optimization and adjustment of TBM tunneling parameters. As a result, data cleaning, distribution statistics, and intelligent prediction are used to establish a prediction and identification method of surrounding rock grades based on self-organizing maps and least squares support vector machine(SVM). The following are the main conclusions reached:(1) A single complete TBM excavation cycle can be divided into section bored without force, rising section, and stable section, and each tunneling parameter approximately obeys the normal distribution relationship.(2) The field penetration index(FPI) and cutterhead torque penetration index(TPI) can indicate the difficulty of tunnel rock excavation. The FPT and TPI are approximately linear to the grade of surrounding rock, which can be used as the sensitive factors of rock machines to inversely predict and identify the grade of surrounding rock.(3) The preprocessing of sample points of interference anomaly data affects the convergence center and fluctuation radius of the surrounding rock prediction model. The ability to accurately predict and identify surrounding rock grades depends on data preprocessing.(4) Different SVM kernel functions have a large impact on surrounding rock grade prediction and identification according to standard test data samples and engineering data. The comprehensive identification rates of surrounding rocks using linear, polynomial, and Gaussian radial basis function kernels are 70.8%, 81.2%, and 87.6%, respectively. The surrounding rock grade prediction and identification model has high predictive accuracy and robustness.