机器学习在 TBM 掘进参数与围岩等级预测中的应用——以引绰济辽工程为例
丁巍 , 魏师琳 , 贾穗子 , 李一丹 , 楚一帆 , 张效忠
有色金属(矿山部分) ›› 2026, Vol. 78 ›› Issue (6) : 75 -87.
机器学习在 TBM 掘进参数与围岩等级预测中的应用——以引绰济辽工程为例
Application of machine learning in predicting TBM tunneling parameters and surrounding rock classification:a case study of the Yinchuojiliao water diversion project
全断面隧道掘进机(TBM)施工过程中产生大量可反映地质状况的实时数据。为充分利用该数据并建立精确的掘进参数与围岩等级预测模型,本文提出了一种基于滑动均值滤波与一阶差分符号变化的掘进段自动分割方法,将掘进循环段划分为空推段、上升段与稳定段;并采用BP、SVM、GPR、RF 和RBF 五种机器学习方法进行对比建模。结果表明:GPR 在掘进参数回归预测中表现最优,基于上升段及进入稳定段60 s 内的数据,稳定段平均推力与单盘扭矩的测试集决定系数R2 分别达到0. 897 与0. 867;RF 在围岩等级分类中精度最高(ACC=0. 985,F1 均值=0. 985),GPR 次之(ACC=0. 930,F1 均值=0. 930)。综合考虑预测精度与模型统一性,本文选取GPR 作为统一预测模型,该模型在提供高精度掘进参数预测值的同时可输出预测置信区间与围岩等级预测概率,为TBM 操作人员提供地质- 机械联合决策支持,有效避免了多模型并行部署带来的工作量繁重与决策冲突等问题。
A large amount of real-time data reflecting geological conditions is generated during the operation of Tunnel Boring Machine (TBM). To fully utilize this data and establish accurate prediction models for TBM tunneling parameters and surrounding rock classification,this paper proposes an automatic segmentation method for tunneling sections based on sliding mean filtering and first-order difference sign changes,dividing the tunneling cycle into air pushing section,rising section and stable section. Five machine learning methods (BP,SVM,GPR,RF and RBF) are compared for modeling. The results show that GPR performs best in the regression prediction of tunneling parameters,with the test set coefficients of determination (R2) for average thrust and single-disk torque reaching 0. 897 and 0. 867,respectively,based on data from the rising section and the first 60 s of the stable section. RF achieves the highest accuracy in surrounding rock classification (ACC=0. 985,F1 mean=0. 985),followed by GPR (ACC=0. 930,F1 mean=0. 930). Considering both prediction accuracy and model uniformity,GPR is selected as the unified prediction model,which can provide not only high-precision tunneling parameter predictions but also prediction confidence intervals and surrounding rock classification probabilities,offering geomechanical joint decision support for TBM operators and effectively avoiding the heavy workload and decision conflicts caused by multi-model parallel deployment.
| [1] |
荆留杰,张娜,杨晨. TBM 及其施工技术在中国的发展趋势[J]. 隧道建设,2016,36(3):331-337. |
| [2] |
|
| [3] |
|
| [4] |
|
| [5] |
刘泉声,潘玉丛,余宏淦, |
| [6] |
|
| [7] |
黄登侠,周小龙,赵健, |
| [8] |
|
| [9] |
|
| [10] |
周小雄,龚秋明,殷丽君, |
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
JTJ 026— 1990 公路隧道设计规范[S]. |
| [15] |
JTJ 026— 1990 Design Specification for Highway Tunnel[S]. |
| [16] |
|
| [17] |
王双敬,王玉杰,李旭, |
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
Kohavi R. A study of cross—validation and bootstrap for accuracy estimation and model selection[C]//Proceedings of the 14th International Joint Conference on Artificial Intelligence—Volume 2. Montreal,Quebec,Canada:Morgan Kaufmann Publishers Inc.,1995:1137-1143. |
| [22] |
|
| [23] |
赵光祖,王亚旭,李尧, |
| [24] |
|
| [25] |
Cortes C, |
| [26] |
李建斌,武颖莹,李鹏宇, |
| [27] |
|
| [28] |
Seeger M. Gaussian processes for machine learning[J]. International Journal of Neural Systems,2004,14(2):69-106. |
| [29] |
|
| [30] |
Breiman L. Random forests[J]. Machine Learning,2001,45(1):5-32. |
| [31] |
|
| [32] |
|
| [33] |
李玉选,郭红斌,黄丹, |
| [34] |
|
| [35] |
王照亚. 金属矿山超千米深TBM 主机组装硐室破坏形式分析与支护设计研究[J]. 有色金属(矿山部分),2023,75(5):26-35. |
| [36] |
|
贵州省科技计划项目(黔科合基础MS〔2025〕244)
贵州省科技厅计划项目(青年引导项目)(黔科合基础QN〔2026〕014-6)
/
| 〈 |
|
〉 |