CN-121996953-A - Coarse-grained soil shear strength prediction method based on deep learning

CN121996953ACN 121996953 ACN121996953 ACN 121996953ACN-121996953-A

Abstract

The invention discloses a coarse-grained soil shear strength prediction method based on deep learning, which comprises the steps of constructing a standardized dataset, constructing a Seft-Net prediction model, dividing the standardized dataset into a training set and a testing set, inputting the training set into the Seft-Net prediction model for training to obtain a trained Seft-Net prediction model, inputting the testing set into the trained Seft-Net prediction model for testing to obtain a tested Seft-Net prediction model, inputting key parameters of coarse-grained soil to be predicted into the tested Seft-Net prediction model, outputting a shear strength peak value q peak , and obtaining shear strength indexes phi 0 and delta phi of coarse-grained soil according to the shear strength peak value q peak . The method solves the problems of low learning efficiency and low prediction accuracy of the existing deep learning method.

Inventors

LI DUO
ZHANG YONGLE
LI YANLONG
SHE LEI
CHEN JUNHAO
SHI ZIYANG

Assignees

西安理工大学

Dates

Publication Date: 20260508
Application Date: 20260107

Claims (9)

1. The coarse-grained soil shear strength prediction method based on deep learning is characterized by comprising the following specific steps of: step 1, constructing a standardized data set; Step 2, constructing Seft-Net prediction models; step 3, dividing the standardized data set into a training set and a testing set, inputting the training set into a Seft-Net prediction model for training to obtain a trained Seft-Net prediction model, and inputting the testing set into a trained Seft-Net prediction model for testing to obtain a Seft-Net prediction model after testing; step 4, inputting key parameters of coarse-grained soil to be predicted into a Seft-Net prediction model after testing, and outputting a shear strength peak value q peak ; And 5, obtaining shear strength indexes phi 0 and delta phi of coarse-grained soil according to the shear strength peak value q peak .
2. The deep learning-based coarse soil shear strength prediction method of claim 1, wherein in step 1, each coarse soil sample in the normalized dataset comprises 13 input features and 1 output feature; The input characteristics comprise uniaxial compressive strength sigma r of the parent rock, particle shape regularity rho, soil particle specific gravity G s , initial pore ratio e 0 , non-uniformity coefficient C u , curvature coefficient C c , confining pressure sigma 3 and typical particle size D max 、D i , wherein D max is the maximum particle size, D represents the particle size, i represents the mass accumulation percentage of soil particles smaller than the particle size D, and i=10, 30, 50, 60 and 80; The rock matrix uniaxial compressive strength sigma r , the particle shape regularity rho and the soil particle specific gravity G s are particle attribute parameters, the initial pore ratio e 0 , the typical particle size D max 、D i , the non-uniformity coefficient C u and the curvature coefficient C c are soil attribute parameters, and the confining pressure sigma 3 is an external factor; The output is characterized by a shear strength peak q peak .
3. The coarse-grained soil shear strength prediction method based on deep learning according to claim 1, wherein in the step 2, the Seft-Net prediction model consists of an input module, a feature transformation module, a feature extraction module, a regression head module and an output module.
4. The method for predicting the shear strength of the coarse-grained soil based on deep learning as set forth in claim 3, wherein the specific processing procedure of the input module is that 13 input features of each coarse-grained soil sample are formed into a 1X 13 one-dimensional array, and the obtained one-dimensional array is converted into a 3X 13 two-dimensional matrix; The input characteristic composition 1×13 one-dimensional array is in the order of uniaxial compressive strength sigma r of the particle attribute parameter parent rock, particle shape regularity rho, soil particle specific gravity G s , initial pore ratio e 0 of the soil mass attribute parameter, typical particle diameter D max 、D 80 、D 60 、D 50 、D 30 、D 10 , non-uniformity coefficient C u , curvature coefficient C c and external factor confining pressure sigma 3 .
5. The coarse-grained soil shear strength prediction method based on deep learning according to claim 4, wherein the specific processing process of the characteristic transformation module is that a two-dimensional matrix is divided into three groups according to particle attribute parameters, soil attribute parameters and external factors, the first Shortcut processing is carried out on the group of particle attribute parameters to obtain a result A, the group of soil attribute parameters is input into a convolution layer to be processed to obtain a result B, the second Shortcut processing is carried out on the group of external factors to obtain a result C, and the result A, the result B and the result C are spliced to obtain a result D.
6. The coarse soil shear strength prediction method based on deep learning according to claim 5, wherein the feature extraction module consists of an inlet flow, 8 intermediate flows, and an outlet flow; The inlet stream sequentially comprises a first convolution layer, a first ReLU activation function, a second convolution layer, a first depth separable convolution module, a second depth separable convolution module and a third depth separable convolution module, and a result D is used as an input of the first convolution layer; The first depth separable convolution module consists of a second ReLU activation function, a first depth convolution layer, a third ReLU activation function, a second depth convolution layer, a first maximum pooling layer and a first point-by-point convolution layer, wherein the output of the second convolution layer is used as the input of the second ReLU activation function and the first point-by-point convolution layer, and the output of the maximum pooling layer and the output of the first point-by-point convolution layer are subjected to first concat operation; The second depth separable convolution module consists of a fourth ReLU activation function, a third depth convolution layer, a fifth ReLU activation function, a fourth depth convolution layer, a second maximum pooling layer and a second point-by-point convolution layer, wherein the result of the first concat operation is used as the input of the fourth ReLU activation function and the second point-by-point convolution layer, and the output of the second maximum pooling layer and the output of the second point-by-point convolution layer are subjected to the second concat operation; The third depth separable convolution module consists of a sixth ReLU activation function, a fifth depth convolution layer, a seventh ReLU activation function, a sixth depth convolution layer, a third maximum pooling layer and a third point-by-point convolution layer, wherein the result of the second concat operation is used as the input of the sixth ReLU activation function and the third point-by-point convolution layer, and the output of the third maximum pooling layer and the output of the third point-by-point convolution layer are subjected to the third concat operation; Each intermediate stream is composed of a seventh depth convolution layer, an eighth depth convolution layer, a ninth depth convolution layer and a fourth concat operation; for the second to eighth intermediate streams, the fourth concat operation result of the last intermediate stream is taken as the input of the seventh depth convolution layer of the current intermediate stream, and the output of the ninth depth convolution layer of the current intermediate stream and the fourth concat operation result of the last intermediate stream are subjected to the fourth concat operation of the current intermediate stream; The outlet flow sequentially comprises a tenth depth convolution layer, an eleventh depth convolution layer, a fourth maximum pooling layer, a fifth concat operation, a twelfth depth convolution layer, a thirteenth depth convolution layer and a global average pooling layer, and the fourth concat operation result of the eighth intermediate flow and the output result of the fourth maximum pooling layer are subjected to the fifth concat operation.
7. The deep learning-based coarse soil shear strength prediction method of claim 5, wherein the regression head module is composed of a first fully connected layer, an eighth ReLU activation function, a second fully connected layer, and the output of the feature extraction module is used as the input of the first fully connected layer.
8. The coarse-grained soil shear strength prediction method based on deep learning according to claim 1, wherein in the step 3, an average absolute error and a mean square error are adopted as a loss function in the training process.
9. The coarse-grained soil shear strength prediction method based on deep learning according to claim 1, wherein the specific process of the step 5 is as follows: Step 5.1, drawing a molar circle and an intensity envelope according to each shear strength peak value q peak and the corresponding confining pressure predicted in the step 4, and obtaining a corresponding internal friction angle phi under each confining pressure according to the intensity envelope; step 5.2, since the coarse-grained soil strength envelope is nonlinear, is expressed as: (1) Wherein phi represents an internal friction angle, p a represents atmospheric pressure, sigma 3 represents confining pressure, phi 0 and delta phi represent intercept and slope of an approximate straight line in logarithmic coordinates when the internal friction angle changes with confining pressure, respectively; and 5.2, drawing a relation diagram of phi and sigma 3 /p a according to the internal friction angle phi obtained in the step 5.1, and obtaining shear strength indexes phi 0 and delta phi of coarse-grained soil according to the relation diagram.

Description

Coarse-grained soil shear strength prediction method based on deep learning Technical Field The invention belongs to the technical field of geotechnical mechanical parameter prediction methods, and particularly relates to a coarse-grained soil shear strength prediction method based on deep learning. Background The coarse granules have the advantages of good compactibility, strong water permeability, convenient material taking and the like, and are widely applied to large geotechnical structures such as earth and rockfill dams, side slopes, revetments, roadbeds and the like, and the mechanical properties of the coarse granules directly influence the safety and stability of engineering. The shear strength of coarse-grained soil is a key parameter for representing the mechanical behavior of the coarse-grained soil, is directly related to the maximum capability of resisting shear damage when the coarse-grained soil is sheared, is crucial for evaluating the stability and bearing capacity of the coarse-grained soil related engineering, and is an indispensable strength parameter in engineering design and construction decisions. Conventionally, shear strength measurement of coarse-grained soil is mainly achieved through a large triaxial test, a direct shear test and other indoor test methods. While these methods can provide accurate test results, the time consuming and costly process limits their feasibility in rapid evaluation and large-scale engineering applications. With the development of data science and artificial intelligence technology, the machine learning method has great application potential in the civil field, and solves the problems of high cost and long time consumption of the traditional test method. Some researches try to apply traditional machine learning models, including mechanical parameters predicted by shallow machine learning technologies such as random forest RF, support vector machines SVM, artificial neural networks and the like, so as to model and predict a small-scale data set. However, when data with complex nonlinear coupling relationships between parameters is processed, shallow machine learning methods generally rely on complex feature engineering to preprocess input parameters, and it is difficult to deeply mine complex patterns hidden in the data. In contrast, deep learning can automatically extract features, has strong nonlinear mapping capability, can effectively capture nonlinear relations in data, and has significant advantages when processing large-scale data. However, the direct application of the general deep learning model to coarse-grained soil shear strength prediction still faces challenges in that the physical meaning and relevance of input parameters (such as particle size, pore ratio, confining pressure, etc.) are different, and the model structure may not be capable of effectively capturing the inherent relationship between the specific physical mechanism and strength parameters of the soil material, resulting in lower prediction accuracy. Disclosure of Invention The invention aims to provide a coarse-grained soil shear strength prediction method based on deep learning, which solves the problem of lower prediction accuracy in the existing deep learning method. The invention adopts the technical scheme that the coarse-grained soil shear strength prediction method based on deep learning comprises the following specific steps: step 1, constructing a standardized data set; Step 2, constructing Seft-Net prediction models; step 3, dividing the standardized data set into a training set and a testing set, inputting the training set into a Seft-Net prediction model for training to obtain a trained Seft-Net prediction model, and inputting the testing set into a trained Seft-Net prediction model for testing to obtain a Seft-Net prediction model after testing; step 4, inputting key parameters of coarse-grained soil to be predicted into a Seft-Net prediction model after testing, and outputting a shear strength peak value q peak; And 5, obtaining shear strength indexes phi 0 and delta phi of coarse-grained soil according to the shear strength peak value q peak. The invention is also characterized in that: In step 1, each coarse soil sample in the standardized dataset contains 13 input features and 1 output feature; The input characteristics comprise uniaxial compressive strength sigma r of the parent rock, particle shape regularity rho, soil particle specific gravity G s, initial pore ratio e 0, non-uniformity coefficient C u, curvature coefficient C c, confining pressure sigma 3 and typical particle size D max、Di, wherein D max is the maximum particle size, D represents the particle size, i represents the mass accumulation percentage of soil particles smaller than the particle size D, and i=10, 30, 50, 60 and 80; The rock matrix uniaxial compressive strength sigma r, the particle shape regularity rho and the soil particle specific gravity G s are particle attribute parameters, t