Robust in-situ stress inversion in an underground powerhouse using tensor synthesis and surrogate-assisted...

Area of Science:

• Geotechnical Engineering
• Rock Mechanics
• Computational Geoscience

Background:

• Accurate in-situ stress field characterization is crucial for deep underground engineering projects.
• Field stress measurements are often scattered, and 3D inversion is computationally intensive.

Purpose of the Study:

• Develop a robust and efficient inversion framework for stress tensor reconstruction.
• Improve the reliability of input data and efficiency of field-scale stress reconstruction.

Main Methods:

• Established a 3D borehole stress synthesis method using particle swarm optimization, Huber loss, Levenberg-Marquardt iteration, and regularization.
• Constructed a surrogate-assisted differential evolution workflow with a radial basis function network and active learning.

Main Results:

• Reduced mean relative error from 14.12% to 9.34%.
• Decreased deviation variance from 2.98 to 1.33.
• Enhanced data reliability and computational efficiency for stress reconstruction.

Conclusions:

• The proposed framework offers a practical solution for stability evaluation and surrounding rock control in deep, complex underground caverns.
• The method effectively denoises measurements, suppresses outliers, and alleviates ill-conditioning.