Deep valley stress field inversion method based on neural network and numerical model

By combining neural networks and numerical models, embedding physical constraints, and performing batch forward modeling, the problems of data sparsity and geological complexity in the inversion of stress fields in deep valleys have been solved, achieving efficient and reliable full-field stress reconstruction and providing a reliable mechanical basis for slope stability evaluation in high mountain and canyon areas.

CN122113606APending Publication Date: 2026-05-29CHANGJIANG INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHANGJIANG INST OF TECH
Filing Date
2026-02-07
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing methods for inverting the geostress field in deeply incised valleys are insufficient to simultaneously guarantee the physical rationality, computational efficiency, and engineering reliability of the inversion results when data is sparse and geological conditions are complex. This results in deficiencies in the evaluation of slope stability and deformation prediction in high mountain and canyon areas.

Method used

By combining neural networks and numerical models, and by embedding physical constraints of equilibrium differential equations and compatibility equations, parameter samples are generated using orthogonal experimental design and batch numerical forward modeling is performed to learn complex nonlinear mapping relationships. Combined with numerical models, the process of valley erosion evolution and stress redistribution is simulated to reconstruct the continuous stress distribution across the entire field.

Benefits of technology

In the case of sparse and limited measurement data, this method can efficiently and reliably reconstruct the continuous stress distribution across the entire valley area, providing a solid mechanical foundation for slope stability evaluation and deformation prediction in high mountain and canyon areas, and solving the problem of complex calculations and low efficiency in existing methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

The application discloses a deep-cut valley stress field inversion method based on a neural network and a numerical model, and comprises the following steps: deep-cut valley stress field inversion region determination and numerical model construction; stress interval estimation, stress boundary parameter sample generation; batch forward calculation, measurement point stress-boundary stress data set construction; BP neural network construction for inversely deducing stress boundary conditions from geostress measurement point data; actual stress boundary conditions are inversely deduced from the measured geostress of the measurement points, and the final geostress field is calculated by combining the physical information neural network and the numerical model in a forward direction. The application embeds physical constraints such as balance differential equations and compatible equations into the loss function of the physical information neural network, solves the problem that the existing method is biased towards data statistics and the physical mechanism is unclear, realizes the physical rationality of the inversion result, generates parameter samples through orthogonal test design and carries out batch numerical forward calculation, and fully covers the parameter space under the condition that the measurement point data is sparse and the distribution is limited.
Need to check novelty before this filing date? Find Prior Art