基于VMD与加权RF的TBM掘进速度预测SHAP解释模型

Shapley Additive Explanations Interpretation Model for Penetration Rate Prediction of Tunnel Boring Machines Based on Variational Mode Decomposition and Weighted Random Forest

为较准确地实现TBM掘进速度(PR)的预测,构建一套基于加权随机森林(RF)结合变分模态分解(VMD)的集成学习预测模型。模型建立过程中,通过收集来自KS隧洞与兰州水源地输水隧洞中涵盖不同岩性下的数据,利用VMD对数据进行4次模态分解,在保留数据特性的同时去除最高频噪音;采用SHAP对未加权传统RF从模型贡献角度进行特征度量,以此实现未加权传统RF加权,并使用RFECV与网格搜索对加权RF进行特征遴选、超参数优化;通过实际工程对模型的性能进行验证,基于SHAP理论对模型从全局与局部进行解释。结果表明:1)所建模型预测精度较高,其在测试集上的均方根误差(MSE)、平均绝对误差(MAE)与决定系数(R^(2))分别为0.0649(m/h)^(2)、0.1875 m/h、0.9254。2)在实际工程的验证中,模型取得了MSE=0.0503(m/h)^(2)、MAE=0.1613 m/h、R^(2)=0.9505的性能表现,精度理想,且性能均高于常用的深度神经网络、支持向量回归、未加权传统RF。3)经过VMD处理可有效提升PR的预测精度,处理后的模型在测试集上MSE、MAE、R^(2)分别提升了82.50%、59.00%、33.25%。4)岩石单轴抗压强度是精准预测PR时最重要的因素,地质参数在预测中的交互性明显优于掘进参数。预测分析重要洞段的PR时,需结合全局与局部2个角度进行综合分析。

A predictive model integrating weighted random forest(RF)and variational mode decomposition(VMD)is constructed to accurately predict the penetration rate(PR)of tunnel boring machines(TBMs).Data from the KS tunnel and the water conveyance tunnel in the Lanzhou water source area,encompassing various lithologies,are collected and decomposed four times using VMD to eliminate high-frequency noise while preserving data characteristics.The traditional RF model′s feature contributions are assessed using Shapley additive explanations(SHAP),facilitating the application of weighted RF.Feature selection and hyperparameter optimization for the weighted RF are achieved through recursive feature elimination with cross-validation and grid search.The model′s performance is validated in practical engineering settings,and it is explained both globally and locally based on SHAP theory.The results indicate:(1)High prediction accuracy with mean square error(MSE),mean absolute error(MAE),and determination coefficient(R^(2))of 0.0649(m/h)^(2),0.1875 m/h,and 0.9254,respectively.(2)In practical engineering applications,the MSE,MAE,and R^(2)are 0.0503(m/h)^(2),0.1613 m/h,and 0.9505,respectively,demonstrating superior accuracy and performance compared to commonly-used deep neural networks,support vector regression,and unweighted RF.(3)VMD processing enhances PR prediction accuracy,showing improvements in MSE,MAE,and R^(2)by 82.50%,59.00%,and 33.25%,respectively.(4)The uniaxial compressive strength of rock is crucial for accurate PR prediction,and the interaction of geological parameters significantly outperforms that of tunneling parameters.For comprehensive analysis in critical tunnel sections,both global and local perspectives should be considered.