Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Civil and Environmental Engineering | en_US |
| dc.contributor.advisor | Yin, Zhen-yu (CEE) | en_US |
| dc.creator | Qiu, Yashi | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/14752 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Machine learning based geological modelling and application | en_US |
| dcterms.abstract | In recent years, with the rapid development of China's economy and the advancement of modernization construction, the demand for underground space development has been increasing day by day, which has also attracted more and more attention from scholars to the problem of geological profile prediction. Accurate geological profile prediction can reveal the distribution characteristics of soil types and strata in a site, which has a significant impact on the design and construction in geotechnical engineering structures. However, the uncertainty and inherent variability of geological formations make accurate geological profile prediction based on sparse borehole data a highly challenging problem. | en_US |
| dcterms.abstract | This study proposes two machine learning algorithms based on sparse boreholes for accurate prediction of geological profiles, each with its own advantages. These algorithms are applied to the prediction of geological profiles in multiple practical engineering projects, obtaining accurate geological profiles to guide engineering practice. Comparing the predictions of the proposed algorithm with those from other machine learning methods to demonstrate its superior performance, particularly in significantly improving boundary accuracy. Furthermore, by evaluating the differences between simulated and monitored surface settlements outside the foundation pit by comparing a linearly interpolated geological model with a machine learning-derived nonlinear profile, both based on sparse borehole data. The actual monitoring data shows that the machine learning-predicted nonlinear geology leads to more accurate finite element simulations. | en_US |
| dcterms.abstract | The main contributions and innovative achievements of this study are as follows: | en_US |
| dcterms.abstract | (1) The MSRF-XGBoost algorithm was proposed, which includes two parts: classification module and filtering module. The classification module is based on the traditional XGBoost algorithm and adds multi receptive field sampling to learn spatial features of different scales between sparse boreholes, enabling accurate prediction of geological profiles through sparse equidistant boreholes alone; The filtering module automatically selects the optimal prediction result from different results predicted by spatial features at different scales through Gaussian filtering, boundary similarity comparison, and other steps. This algorithm solves the dependence of existing machine learning algorithms on training profiles and improves the boundary accuracy of prediction results by about 40% in nonlinear geological profile prediction. | en_US |
| dcterms.abstract | (2) An improved GCN algorithm has been proposed. By improving and optimizing the feature expression, adjacency matrix design, and model training method of the GCN algorithm, the improved GCN algorithm can predict the soil type of unknown points through their coordinates, and thus predict the complete geological profile, solving the problem of existing machine learning algorithms relying on equidistant drilling for training. This algorithm can predict accurate geological profiles through sparse non equidistant boreholes. Compared with other machine learning methods, the boundary accuracy of nonlinear geological profile prediction has been improved by about 30%. | en_US |
| dcterms.abstract | (3) The two proposed machine learning methods are applied to multiple practical engineering projects. Validation through borehole accuracy assessment of verification boreholes shows prediction accuracy between 80%-99% across three different prediction profiles. Comparative analysis with other machine learning algorithms confirms the proposed methods' advantages in reducing training data requirements while improving prediction accuracy. | en_US |
| dcterms.abstract | (4) Taking a foundation pit in Hangzhou as an example, linear geological profiles based on sparse boreholes linear interpolation and nonlinear geological profile finite element models based on algorithm prediction are established. During the simulation of excavation process, the non-linear geological profile finite element model showed a difference of 2% between the surface settlement value outside the pit and the actual monitored settlement value, while the linear geological profile finite element model simulation result differed from the actual settlement value by 12%, indicating that accurate non-linear geological profiles can make finite element model simulation more accurate. | en_US |
| dcterms.extent | xv, 113 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2026 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14752