当前位置: X-MOL 学术Tunn. Undergr. Space Technol. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Predicting EPBM advance rate performance using support vector regression modeling
Tunnelling and Underground Space Technology ( IF 6.7 ) Pub Date : 2020-10-01 , DOI: 10.1016/j.tust.2020.103520
Soroush Mokhtari , Michael A. Mooney

Abstract Earth pressure balance shield tunnel boring machines (EPBM) are widely used in tunneling practice yet the mechanics that define ground-EPBM interaction, specifically the advance rate, are not well understood. In the study presented here, machine learning techniques including feature selection, support vector regression (SVR) and partial dependence plots (PDP), were successfully applied to EPBM data to develop and explain EPBM advance rate modeling through five widely varying soil types. The geotechnical conditions were implicitly incorporated into the analysis by developing soil formation-specific SVR models. The SVR models were highly successful in capturing AR behavior, exhibiting R2 values of 0.88–0.95 when independently evaluated with test data. Automatic feature selection revealed the same EPBM parameters of notable influence on AR across all ESUs, including net thrust, cutterhead torque, foam flow rate and screw conveyor torque. The SVR models, however, revealed considerably different relationships between these key parameters and AR, indicating that the soil plays a significant role in AR behavior. PDP analysis captured the sensitivity of AR to each key parameter as a function of parameter magnitude. The PDP results show that AR is positively correlated (increasing AR with increasing parameter value) and/or negatively correlated (decreasing AR with increasing parameter value) to varying degrees as a function of parameter value, all of which is strongly soil dependent.

中文翻译:

使用支持向量回归模型预测 EPBM 提前率性能

摘要 土压平衡盾构隧道掘进机(EPBM)广泛用于隧道掘进实践,但定义地面-EPBM相互作用的力学,特别是推进率,还不是很清楚。在这里介绍的研究中,包括特征选择、支持向量回归 (SVR) 和部分依赖图 (PDP) 在内的机器学习技术已成功应用于 EPBM 数据,以通过五种广泛变化的土壤类型开发和解释 EPBM 推进率建模。通过开发特定于土壤形成的 SVR 模型,岩土工程条件被隐含地纳入分析。SVR 模型在捕捉 AR 行为方面非常成功,当使用测试数据进行独立评估时,其 R2 值为 0.88-0.95。自动特征选择揭示了对所有 ESU 的 AR 有显着影响的相同 EPBM 参数,包括净推力、刀盘扭矩、泡沫流速和螺旋输送机扭矩。然而,SVR 模型揭示了这些关键参数与 AR 之间相当不同的关系,表明土壤在 AR 行为中起着重要作用。PDP 分析捕获 AR 对每个关键参数的敏感性,作为参数幅度的函数。PDP 结果表明,作为参数值的函数,AR 在不同程度上呈正相关(随着参数值的增加而增加)和/或负相关(随着参数值的增加而减少),所有这些都强烈依赖于土壤。然而,揭示了这些关键参数与 AR 之间存在显着不同的关系,表明土壤在 AR 行为中起着重要作用。PDP 分析捕获 AR 对每个关键参数的敏感性,作为参数幅度的函数。PDP 结果表明,作为参数值的函数,AR 在不同程度上呈正相关(随着参数值的增加而增加)和/或负相关(随着参数值的增加而减少),所有这些都强烈依赖于土壤。然而,揭示了这些关键参数与 AR 之间存在显着不同的关系,表明土壤在 AR 行为中起着重要作用。PDP 分析捕获 AR 对每个关键参数的敏感性,作为参数幅度的函数。PDP 结果表明,作为参数值的函数,AR 在不同程度上呈正相关(随着参数值的增加而增加)和/或负相关(随着参数值的增加而减少),所有这些都强烈依赖于土壤。
更新日期:2020-10-01
down
wechat
bug