Skip to main content
Have a personal or library account? Click to login
Rock mass properties prediction method in TBM tunnel based on principal component analysis (PCA) Cover

Rock mass properties prediction method in TBM tunnel based on principal component analysis (PCA)

Civil and Environmental Engineering
Volume 22 (2026): Issue 2 (June 2026)
By: ,  ,  ,   and    
Open Access
|Jun 2026
Figures & tables

Figures & Tables

Table 1:

Basic information of the datasets and projects (Liu et al. 2019, 2020)

Data sourcesYSYHZJHZQD
Area IArea IIArea IArea IIArea IArea II
Number of samples [-]46080502304625056100
TBM type [-]Open typeDouble-shieldDouble-shieldDouble-shieldDouble-shield
Number of tunneling variables [-]178202276276202
TBM diameter [m]7.97.06.26.07.0
Number of cutters [-]5648424048
Figure 1:

Division of the field datasets

Figure 2:

Part of the covariance matrix results

Figure 3:

Part of the covariance matrix results

Table 2:

Mean absolute percentage error of the sub-models in the 10-fold-corss-validation

OrderTraining sub-setTest sub-setBPNNSVRRF
UCS [%]Jf [%]UCS [%]Jf [%]UCS [%]Jf [%]
12–1019.78.49.29.29.08.9
21, 3–1028.68.29.18.38.98.8
31–2, 4–1039.09.48.58.68.79.0
41–3, 5–1049.69.29.99.68.89.8
51–4, 6–10510.09.78.89.210.19.1
61–5, 7–1068.48.79.28.38.910.1
71–6, 8–1079.78.99.59.69.48.8
81–7, 9–1089.39.58.98.710.69.4
91–8, 1099.79.49.29.89.18.9
101–9109.48.99.29.29.38.8
Figure 4:

Eigenvalues and the accumulated information of the top 10 variable in the four projects

Figure 5:

Comparison between the predict UCS to the actual values in different project

Figure 6:

Comparison between the predict Jf to the actual values in different project

Figure 7:

Predicted results of UCS in test sets

Figure 8:

Predicted results of Jf in test sets

Figure 9:

Comparison between the predicted UCS and Jf by SVR, RF, and the actual values in validation sets

Table 3:

Predicted MAPE values of the validation sets in other four projects

ProjectsUCSJf
BPNN [%]SVR [%]RF [%]BPNN [%]SVR [%]RF [%]
YS10.110.310.78.48.58.8
ZJ11.011.511.48.89.49.1
YH8.18.39.05.36.06.2
HZ9.78.99.69.18.89.9
QD10.510.210.89.78.910.4
Average value9.99.810.38.38.38.9
Figure 10:

Comparison between the predicted UCS and Jf by SVR, RF, and the actual values in test sets

Table 4:

Predicted MAPE values of the test sets in other four projects

ProjectsUCSJf
BPNN [%]SVR [%]RF [%]BPNN [%]SVR [%]RF [%]
YS13.012.814.412.713.214.8
ZJ11.912.213.912.311.814.2
HZ11.912.313.211.011.213.5
Average value12.312.513.812.012.114.2
Table 5:

The network structures and time costs of PCA and FV models

PCA modelFV models
YSYHZJHZQDYSYHZJHZQD
Number of input neurons [-]54344178202276276202
Number of hidden neurons [-]865665664727264
Time costs [s]UCS64444202179261252188
Jf64353196165249266193
Figure 11:

Comparison between the calculating accuracy in validation sets of PCA model and FV models

Figure 12:

Comparison between the calculating accuracy in test sets of PCA model and FV models

DOI: https://doi.org/10.2478/cee-2026-0051 | Journal eISSN: 2199-6512 (formerly 1336-5835) | Journal ISSN: 1336-5835
Journal RSS Feed
Language: English
Page range: 764 - 783
Submitted on: Sep 1, 2025
Accepted on: Oct 23, 2025
Published on: Jun 19, 2026
Published by: University of Žilina
In partnership with: Paradigm Publishing Services
Publication frequency: 4 issues per year
Keywords:
Tunnel boring machine,
Machine learning,
Principal component analysis,
Rock mass parameters
Related subjects:
Business and economics,
Business management,
Industries,
Environmental management

© 2026 Qingting Meng, Hai Wang, Liang Bai, Fei Zhao, Jian Bai, published by University of Žilina
This work is licensed under the Creative Commons Attribution 4.0 License.

Volume 22 (2026): Issue 2 (June 2026)
Previous article
Next article