标题：Improved support vector regression models for predicting rock mass parameters using tunnel boring machine driving data
作者：Liu, Bin; Wang, Ruirui; Guan, Zengda; Li, Jianbin; Xu, Zhenhao; Guo, Xu; Wang, Yaxu
作者机构：[Liu, Bin; Wang, Ruirui; Xu, Zhenhao; Wang, Yaxu] Shandong Univ, Geotech & Struct Engn Res Ctr, Jinan, Shandong, Peoples R China.; [Liu, Bin; Wang, 更多
通讯作者：Guo, Xu;Guo, X
通讯作者地址：[Guo, X]Chinese Univ Hong Kong, Dept Math, Shatin, Room G07,Lady Shaw Bldg, Hong Kong, Peoples R China.
来源：TUNNELLING AND UNDERGROUND SPACE TECHNOLOGY
关键词：Rock mass parameter; Support vector regression; Tunnel boring machine;; TBM driving data
摘要：The sensitivity of tunnel boring machines (TBMs) to complex rock mass parameters makes the accurate and reliable prediction of these parameters crucial for the selection of reasonable operational parameters and the reduction of construction project risks. We introduce and verify a TBM driving data-based method for predicting rock mass parameters including the uniaxial compressive strength (UCS), brittleness index (BI), distance between planes of weakness (DPW), and orientation of discontinuities (alpha). For this purpose, an artificial intelligence (AI) algorithm, namely support vector regression (SVR), is improved by the stacked single-target (SST) technique and used to establish rock mass parameter prediction models. A dataset of 180 samples is established based on parameters from the 4th section of the Water Supply Project from Songhua River, with 150 randomly selected samples used for training. The constructed models are applied to the remaining 30 samples, and the mean squared percentage error (MSPE) of prediction results for UCS, BI, DPW, and alpha are determined as 3.0%, 4.6%, 3.0%, and 2.5%, respectively, while the respective determination coefficients (R-2) are obtained as 0.83, 0.75, 0.63, and 0.63. The above results are better than the results of common SVR method, and show that the developed models can effectively simulate rock mass parameters and their sudden changes, i.e., the prediction of these parameters based on TBM driving data is both feasible and practical. Moreover, the initial models are used on the dataset, the comparison between their results and the results of proposed models verify the positive effect of the SST on SVR method.