Search

CN-122020022-A - Coarse-grained soil dynamic shear modulus and damping ratio prediction method based on attention mechanism

CN122020022ACN 122020022 ACN122020022 ACN 122020022ACN-122020022-A

Abstract

The invention discloses a coarse-grained soil dynamic shear modulus and damping ratio prediction method based on an attention mechanism, which comprises the following specific steps of 1, constructing a coarse-grained soil test data set, 2, constructing a GA-Net prediction model, 3, dividing the data set into a training set and a testing set, inputting the training set into the GA-Net prediction model for training to obtain a trained GA-Net prediction model, testing the trained GA-Net prediction model by using the testing set to obtain a tested GA-Net prediction model, and 4, inputting key parameters of coarse-grained soil to be tested into the tested GA-Net prediction model to obtain a group of dynamic shear modulus G and a group of damping ratio lambda. The method has high prediction precision and high prediction effect.

Inventors

  • LI DUO
  • SHI ZIYANG
  • LI YANLONG
  • BU PENG
  • QIU WEN
  • ZHANG YONGLE

Assignees

  • 西安理工大学

Dates

Publication Date
20260512
Application Date
20251231

Claims (8)

  1. 1. The coarse-grained soil dynamic shear modulus and damping ratio prediction method based on an attention mechanism is characterized by comprising the following specific steps of: step 1, constructing a coarse-grained soil test data set; step 2, constructing a GA-Net prediction model; step 3, dividing the data set constructed in the step 1 into a training set and a testing set, inputting the training set into the GA-Net prediction model for training to obtain a trained GA-Net prediction model, and then testing the trained GA-Net prediction model by using the testing set to obtain a tested GA-Net prediction model; And step 4, inputting key parameters of coarse-grained soil to be tested into the tested GA-Net prediction model to obtain a group of dynamic shear modulus G and a group of damping ratio lambda.
  2. 2. The method for predicting the dynamic shear modulus and damping ratio of coarse-grained soil based on an attention mechanism as set forth in claim 1, wherein in step 1, the coarse-grained soil test dataset comprises 14 key parameters, specifically, the regularity ρ of the particle shape and the uniaxial compressive strength of the parent rock Specific gravity of soil particle Initial void ratio Non-uniformity coefficient Coefficient of curvature C c , effective mean principal stress Dynamic shear strain sequence and characteristic particle size 、 、 、 、 、 ; Wherein, the Representing the maximum particle size of the coarse-grained soil sample; The grain size of the coarse-grained soil sample grain grade distribution curve is smaller than the grain size corresponding to a certain grain size content i, i=80, 60, 50, 30 and 10.
  3. 3. The attention mechanism-based coarse-grained soil dynamic shear modulus and damping ratio prediction method as claimed in claim 2, wherein the dynamic shear strain sequence is composed of a plurality of dynamic shear strains from small to large The value of the dynamic shear strain is 0.00001-0.1.
  4. 4. The attention mechanism-based coarse-grained soil dynamic shear modulus and damping ratio prediction method according to claim 2, wherein in step 2, the GA-Net prediction model consists of an input module, an attention module and a regression head module; The processing process of the input module comprises the steps of forming a sequence A of 1×14 by 14 key parameters of each coarse-grained soil sample, splicing 2 zero values at the tail of the sequence A to form a sequence B of 1×16, mapping each element in the sequence B to a characteristic vector of 1×16 dimensions, and expanding the sequence B into a batch×1×16×16 matrix; The attention module consists of an encoder and a decoder; the regression head module consists of a full connecting layer and outputs a dynamic shear modulus G and a damping ratio lambda.
  5. 5. The method for predicting the coarse-grained soil dynamic shear modulus and damping ratio based on an attention mechanism according to claim 4, wherein the specific processing process of the encoder is that a batch×1×16×16 matrix is input and then is processed by a first multi-head self-attention layer, a result A is output, the result A and the batch×1×16×16 matrix are added and then are processed by a first layer normalization to obtain a result B, the result B is input to a first feedforward network for processing to obtain a result C, and the result B and the result C are added and then are processed by a second layer normalization to obtain a result D.
  6. 6. The attention mechanism-based coarse-grained soil shear modulus and damping ratio prediction method according to claim 4, wherein the specific processing process of the decoder is that after the input of a batch×1×16×16 matrix, the second multi-head self-attention layer processing is performed, a result E is output, the result E and the batch×1×16×16 matrix are spliced and then subjected to the third layer normalization processing, a result F is obtained, the result F and the result D are input into a cross attention layer for processing, a result G is obtained, the result G and the result F are added and then subjected to the fourth layer normalization processing, a result H is obtained, the result H is input into a second feedforward network for processing, a result J is obtained, the result J and the result H are added and subjected to the fifth layer normalization processing, and the result K is used as the input of a regression head module.
  7. 7. The attention mechanism-based coarse-grained soil dynamic shear modulus and damping ratio prediction method of claim 1, wherein in step 3, the dividing ratio of the training set to the test set is 5:1.
  8. 8. The attention mechanism-based coarse-grained soil dynamic shear modulus and damping ratio prediction method according to claim 1, wherein in step 3, a loss function adopted in the training process is: (1) In the formula (1), MAE represents an average absolute error, A i represents an actual value, P i represents a preliminary test, and n represents the number of coarse-grained soil samples; (2) in formula (2), RMSE represents root mean square error; (3) in formula (3), MAPE represents the average absolute percentage error; (4) In the formula (4), R 2 represents a determination coefficient, and Ā represents an average value of actual values.

Description

Coarse-grained soil dynamic shear modulus and damping ratio prediction method based on attention mechanism Technical Field The invention belongs to the technical field of geotechnical mechanical parameter prediction methods, and particularly relates to a coarse-grained soil dynamic shear modulus and damping ratio prediction method based on an attention mechanism. Background Coarse-grained soil is a non-cohesive mixed soil which consists of stone blocks, broken stones, pebbles or stone chips and contains a large amount of coarse particles, and is widely applied to projects such as water conservancy, civil engineering, traffic and the like. Coarse-grained soil is used as a main filling material of large-scale geotechnical structures such as earth-rock dams and the like, and the dynamic characteristics of the coarse-grained soil have important influence on earthquake response analysis, earthquake stability evaluation in geotechnical structures and earthquake-resistant design. Dynamic shear modulus (G) and damping ratio (lambda) are key constitutive parameters of soil dynamic response, and directly influence the accuracy of engineering earthquake-resistant design, numerical simulation analysis and earthquake response evaluation. In recent years, the development of artificial intelligence technology has promoted the application of machine learning technology in prediction of soil mechanics parameters. For determining two key parameters of dynamic shear modulus and damping ratio in coarse-grained soil dynamic characteristics, shallow machine learning, such as a support vector machine, is mainly adoptedSVM) random forestRF), fully connected neural networkFCN) and the like to predict the maximum shear modulus or shear wave velocity at low strain conditions. However, these methods have a problem of low prediction accuracy when dealing with highly nonlinear and multivariable coupling problems such as dynamic shear modulus decay and damping ratio variation. Therefore, a method capable of accurately and efficiently predicting the dynamic shear modulus and damping ratio of coarse-grained soil is needed. Disclosure of Invention The invention aims to provide a coarse-grained soil dynamic shear modulus and damping ratio prediction method based on an attention mechanism, and solves the problem of low prediction accuracy in the existing method. The invention adopts the technical scheme that the coarse-grained soil dynamic shear modulus and damping ratio prediction method based on an attention mechanism comprises the following specific steps: step 1, constructing a coarse-grained soil test data set; step 2, constructing a GA-Net prediction model; step 3, dividing the data set constructed in the step 1 into a training set and a testing set, inputting the training set into the GA-Net prediction model for training to obtain a trained GA-Net prediction model, and then testing the trained GA-Net prediction model by using the testing set to obtain a tested GA-Net prediction model; And step 4, inputting key parameters of coarse-grained soil to be detected into the tested GA-Net prediction model to obtain a group of dynamic shear modulus G and a group of damping ratio lambda, and calculating constitutive parameters. The invention is also characterized in that: in the step 1, the coarse-grained soil test data set comprises 14 key parameters, specifically, the particle shape regularity rho and the uniaxial compressive strength of the parent rock Specific gravity of soil particleInitial void ratioNon-uniformity coefficientCoefficient of curvature C c, effective mean principal stressDynamic shear strain sequence and characteristic particle size、、、、、; Wherein, the Representing the maximum particle size of the coarse-grained soil sample; The grain size of the coarse-grained soil sample grain grade distribution curve is smaller than the grain size corresponding to a certain grain size content i, i=80, 60, 50, 30 and 10. The dynamic shear strain sequence is composed of a plurality of dynamic shear strains from small to largeThe value of the dynamic shear strain is 0.00001-0.1. In the step 2, the GA-Net prediction model consists of an input module, an attention module and a regression head module; The processing process of the input module comprises the steps of forming a sequence A of 1×14 by 14 key parameters of each coarse-grained soil sample, splicing 2 zero values at the tail of the sequence A to form a sequence B of 1×16, mapping each element in the sequence B to a characteristic vector of 1×16 dimensions, and expanding the sequence B into a batch×1×16×16 matrix; The attention module consists of an encoder and a decoder; the regression head module consists of a full connecting layer and outputs a dynamic shear modulus G and a damping ratio lambda. The specific processing procedure of the encoder is that a batch multiplied by 1 multiplied by 16 matrix is input and then is processed by a first multi-head self-attention layer, a result A i