Rock-soil stress-strain method based on Bi-LSTM fused with CBAM deep learning

Through Bi-LSTM fusion CBAM deep learning method, the idealized assumption deviation and parameter calibration difficulties of constitutive models in geotechnical engineering are solved, and more accurate nonlinear stress and strain description of soil bodies and time-related characteristics are achieved, which improves the Model performance and universality.

CN119047333APending Publication Date: 2024-11-29LANZHOU UNIV
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411291453.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-14
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

In existing geotechnical engineering, constitutive models have problems such as idealized assumptions deviating from reality when describing the nonlinear mechanical behavior of soils, difficulty in parameter calibration, and failure to consider time-related characteristics, resulting in insufficient universality and accuracy of the model.

Method used

Bi-LSTM fusion CBAM deep learning method is used to obtain stress and strain data through triaxial compression test, use bidirectional LSTM to process time-related information, and optimize the model through CBAM attention mechanism, and train it with Adam optimizer to overcome the hypothetical deviation of traditional models. and parameter calibration is difficult.

Benefits of technology

It realizes a more accurate description of the nonlinear stress and strain behavior of soil, improves model performance and universality, overcomes the assumption deviation and parameter calibration difficulties of traditional models, and can effectively deal with time-related characteristics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047333A_ABST
    Figure CN119047333A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of rock and soil, and discloses a Bi-LSTM-fused CBAM deep learning rock and soil stress strain method based on Bi-LSTM. According to the deep learning rock-soil stress-strain method based on Bi-LSTM fused with CBAM, weight parameters and bias parameters in a model can be effectively optimized by applying an Adam optimizer, so that the purpose of minimizing loss is achieved, meanwhile, an effective model can be obtained by training the model, the ability of the models is evaluated by utilizing the models, and the method has the advantages of being simple in structure, convenient to operate and high in practicability. Compared with other models, the method can overcome the defects that a theoretical model adopts multiple assumptions, the universality is poor and advanced model parameters are difficult to calibrate, forward information and backward information can be processed at the same time by processing data information of rock and soil stress and strain through a bidirectional structure, the model performance is greatly improved, and the method is suitable for popularization and application. The method can overcome the defect that the traditional neural network method does not consider the time-dependent characteristics of the stress-strain behavior of the soil body.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of geotechnical technology, and in particular to a geotechnical stress-strain method based on Bi-LSTM fusion CBAM deep learning. Background Art

[0002] Soil is the carrier of geotechnical infrastructure. Determining the nonlinear mechanical response of soil under load conditions is of great significance to the design and construction of infrastructure. Due to the complex internal structure and diverse components of soil, soil has complex nonlinear deformation characteristics under external loads; the mathematical model that currently describes the nonlinear mechanical behavior of soil is called the constitutive model; according to different research methods, soil constitutive models can be divided into traditional theoretical models and neural network constitutive models. Traditional constitutive models usually regard soil as a continuous medium, and based on a variety of idealized assumptions, they approximately describe the nonlinear stress-strain relationship of soil, which more or less deviates from the actual stress-strain behavior of soil. Under specific conditions, they can better describe the specific stress-strain behavior of a certain type of soil. For example, the modified Cambridge model can describe the shear dilatancy of sand, but cannot describe the creep characteristics of clay. In order to enable the model to describe more nonlinear characteristics of soil, the advanced constitutive model adds characteristic parameters to the model. These have no clear physical meaning, which makes parameter calibration difficult and leads to parameter optimization problems. In order to better solve geotechnical engineering problems, the neural network method is a data-driven method that directly extracts the nonlinear stress-strain characteristics of soil from the stress-strain data of soil. The method is simple, has strong universality, and is easy to promote. Summary of the invention

[0003] The present invention provides a geotechnical stress-strain method based on Bi-LSTM fusion CBAM deep learning, which has the advantages of simple method, strong universality and easy promotion, and solves the problems in the above-mentioned background technology.

[0004] The present invention provides the following technical solution: a geotechnical stress-strain method based on Bi-LSTM fusion CBAM deep learning, and the method steps are as follows:

[0005] S1. Establish an original data set and pre-process the data. Use a triaxial compression test to obtain strain data of different soil samples under a specified stress loading step, establish an original data set of stress and strain, and standardize the data. The standardization calculation method formula is as follows:

[0006]

[0007] Among them, x represents the original data, x * represents the standardized data, x min and x max Respectively represent the minimum and maximum values ​​of the feature;

[0008] S2. Establish a deep learning model through Bi-LSTM;

[0009] S3, carrying the attention module through the CBAM attention mechanism;

[0010] S4, model training, the training steps include:

[0011] a. Parameter initialization: Parameters such as weights and learning rates need to be set before they can be correctly applied to the model;

[0012] b. Establish the loss function and set the stopping condition: The loss function is an important learning criterion that can be used to measure the performance of a network and is often related to the optimization amount. The mean square error loss function is used to calculate the mean square error between them. The calculation formula is as follows:

[0013]

[0014] Where N is the length of the input sequence, is the predicted value of the model, y i is the true value;

[0015] c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized;

[0016] d. By carefully adjusting the parameters and functions, an effective network training process can be achieved. That is, according to the pre-set number of training times, the parameters of the model are continuously adjusted until a certain threshold is reached. In this way, even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not been significantly improved, the deep learning model will be terminated.

[0017] Preferably, the Bi-LSTM in S2 uses RNN and its variant LSTM to process data information, which has a bidirectional structure. The basic architecture of the LSTM consists of five parts: input x t 、output h t , Update gate i t 、Forget gate t , output gate o t ,in:

[0018] Forget Gate t The expression of is as follows:

[0019] f t =σ(W f *[h t-1 ,x t ]+b f )

[0020] Update Gate t The expression of is as follows:

[0021] i t =σ(W i *[h t-1 ,x t ]+b i )

[0022] c' t =tanh(W c *[h t-1 ,x t ]+b c )

[0023] c t =f t *c t-1 +i t *c' t

[0024] Output gate o t The expression of is as follows:

[0025] o t =σ(W o *[h t-1 ,x t ]+b o )

[0026] h t =o t *tanh(c t ).

[0027] The present invention has the following beneficial effects:

[0028] 1. This geotechnical stress-strain method based on Bi-LSTM fusion CBAM deep learning can effectively optimize the weight parameters and bias parameters in the model by applying the Adam optimizer, so as to achieve the purpose of minimizing the loss. At the same time, by training the model, an effective model can be obtained, and these models can be used to evaluate their capabilities for comparison with other models. This method can overcome the defects of theoretical models adopting multiple assumptions, poor universality and difficulty in calibrating advanced model parameters.

[0029] 2. The geotechnical stress-strain method based on Bi-LSTM fusion CBAM deep learning can process the forward and backward information at the same time by processing the geotechnical stress-strain data information with a bidirectional structure, which greatly improves the model performance. This method can overcome the defect that the traditional neural network method does not consider the time-related characteristics of soil stress-strain behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a structural block diagram of the LSTM unit of the present invention;

[0031] Figure 2 It is a structural block diagram of the CBAM module of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] See also Figure 1-Figure 2 , see Figure 1-2 , a geotechnical stress-strain method based on Bi-LSTM fusion CBAM deep learning, the method includes the following steps:

[0034] S1. Establish the original data set and preprocess the data. Use triaxial compression test to obtain strain data of different soil samples under specified stress loading steps, establish the original data set of stress and strain, and standardize the data. Data standardization is a key preprocessing process. Its purpose is to convert variables of different dimensions and ranges to the same order of magnitude, so as to ensure that the model can evaluate each feature equally and will not have a disproportionate impact on the results due to some features with large numerical ranges. This helps to improve the convergence speed of the algorithm and ensure that the feature information is effectively utilized. Min-max standardization is adopted. This method is also called normalization. It is performed by scaling the data to the range of [0, 1]. The standardization calculation method formula is as follows:

[0035]

[0036] Among them, x represents the original data, x * represents the standardized data, x min and x max Respectively represent the minimum and maximum values ​​of the feature;

[0037] S2. Establish a deep learning model through Bi-LSTM;

[0038] S3, carrying the attention module through the CBAM attention mechanism;

[0039] S4, model training, the training steps include:

[0040] a. Parameter initialization: Parameters such as weights and learning rates need to be set before they can be correctly applied to the model;

[0041] b. Establish the loss function and set the stop condition: The loss function is an important learning criterion, which can be used to measure the performance of a network and is often related to the optimization amount. The geotechnical stress-strain method of the present invention uses the mean square error loss function to calculate the mean square error between them, so as to better describe the performance of a network and more accurately predict future results to achieve better results. The calculation formula is shown in the following formula:

[0042]

[0043] Where N is the length of the input sequence, is the predicted value of the model, y i is the true value;

[0044] c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized to minimize the loss. The Adam optimizer can automatically adjust the learning rate, so it has become the preferred optimizer in the field of deep learning.

[0045] d. By carefully adjusting the parameters and functions, an effective network training process can be achieved, that is, according to the pre-set number of training times, the parameters of the model are continuously adjusted until a certain threshold is reached. In this way, even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not been significantly improved, the deep learning model will be terminated. By training the model, an effective model can be obtained, and these models can be used to evaluate their capabilities for comparison with other models.

[0046] Among them, Bi-LSTM in S2 uses RNN and its variant LSTM to process data information. It has a bidirectional structure. The basic architecture of LSTM consists of five parts: input x t 、output h t , Update gate i t 、Forget gate t , output gate o t , its core part uses a special gating unit, such as Figure 1 As shown in the figure, c in the figure is the cell state, which will be updated after passing through the update gate and output after the output gate. It can be seen from the figure that the entire LSTM unit structure has traces of the cell state. Therefore, it has a memory capacity that exceeds that of RNN. Even if the data is far apart, it can still be effectively retained, among which:

[0047] Forget Gate tThe expression of is as follows:

[0048] f t =σ(W f *[h t-1 ,x t ]+b f )

[0049] This gate determines whether the information stays or goes away. t-1 and x t After that, a numerical output will be made. The range of the numerical value is 0-1. The actual function of the output value is the amount of information that passes through the forget gate. If no information passes, then the output is 0. If all the information passes, then the output is 1. σ is the sigmoid function, W f is the unit weight matrix, b f is the bias matrix of the unit;

[0050] Update Gate t The expression of is as follows:

[0051] i t =σ(W i *[h t-1 ,x t ]+b i )

[0052] c' t =tanh(W c *[h t-1 ,x t ]+b c )

[0053] The update gate determines what value to update, c' t will also be added to c, which is a vector, and then the cell state is updated. The old cell state is the same as f t Multiply, discard the information that is decided to be discarded, i t *c' t is the new candidate value, which is expressed as follows:

[0054] c t =f t *c t-1 +i t *c' t

[0055] Output gate o t The expression of is as follows:

[0056] o t =σ(W o *[h t-1 ,x t ]+b o )

[0057] h t =o t *tanh(c t )

[0058] It ultimately determines the output value of the unit, and the output gate calculates h t The sigmoid layer determines which part of the cell state is output, and the tanh layer processes it to obtain a value between -1 and 1, and finally determines the output.

[0059] The above content describes the basic structure of a unidirectional LSTM unit, which can only process information before the current moment. In contrast, RNN and its variants (such as bidirectional LSTM) have a bidirectional structure and can process forward information and backward information at the same time, providing a more comprehensive data understanding and analysis, so that each moment can determine the final output based on the previous information and the subsequent information. This bidirectional structure greatly improves the model performance because they can process information at two moments at the same time. The geotechnical stress-strain method adopted in the present invention is based on Bi-LSTM (Bidirectional LSTM), which can capture the information of the previous and next moments, and overcomes the defect that the traditional neural network method does not consider the time-related characteristics of the stress-strain behavior of the soil.

[0060] Among them, the implementation process of the CBAM (Convolutional Block Attention Module) attention mechanism in S3 is as follows:

[0061] Channel attention mechanism and spatial attention mechanism. Hybrid attention refers to combining spatial and channel attention to maximize the weight of task-related features by selectively focusing on different dimensional information of the input feature map. As a typical hybrid attention implementation, CBAM places special emphasis on how to effectively integrate these two attention mechanisms to optimize the performance of the model. Its structural details are shown in the figure below. CBAM has played a huge role in improving the attention mechanism of traditional convolutional neural networks. It takes both spatial attention and channel attention into consideration, allowing the entire model to process information in an effective way. Taking dual attention of space and channel helps to improve the performance of the neural network. The former focuses on the key pixel areas that affect the classification results, while the latter focuses on giving different weights to different channels, which helps to optimize the performance of the model.

[0062] This module consists of two independent parts: one is CAM, which is used to handle channel attention; the other is SAM, which is used to handle spatial attention. Together, they constitute the core functionality of this module.

[0063] CAM considers both draw and maximum pooling to reduce the information loss caused by pooling. SAM first performs global maximum pooling and global average pooling, then performs convolution operations to reduce the channel to 1, and finally uses sigmoid to generate a spatial attention feature map, which is then multiplied with the channel attention feature map.

[0064] The accompanying drawings in the embodiments of the present invention: different types of section lines in the drawings are not marked according to national standards, nor do they impose requirements on the materials of the components. Instead, they are used to distinguish the cross-sectional views of the components in the drawings.

Claims

1. Based on the Bi-LSTM fusion CBAM deep learning geotechnical stress-strain method, the method steps are as follows: S1. Establish an original data set and pre-process the data. Use a triaxial compression test to obtain strain data of different soil samples under a specified stress loading step, establish an original data set of stress and strain, and standardize the data. The standardization calculation method formula is as follows: in, x represents the original data, x * represents the standardized data, x min and x max Respectively represent the minimum and maximum values ​​of the feature; S2. Establish a deep learning model through Bi-LSTM; S3, carrying the attention module through the CBAM attention mechanism; S4, model training, the training steps include: a. Parameter initialization: Parameters such as weights and learning rates need to be set before they can be correctly applied to the model; b. Establish the loss function and set the stopping condition: The loss function is an important learning criterion that can be used to measure the performance of a network and is often related to the optimization amount. The mean square error loss function is used to calculate the mean square error between them. The calculation formula is as follows: Where N is the length of the input sequence, is the predicted value of the model, y i is the true value; c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized; d. By carefully adjusting the parameters and functions, an effective network training process can be achieved. That is, according to the pre-set number of training times, the parameters of the model are continuously adjusted until a certain threshold is reached. In this way, even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not been significantly improved, the deep learning model will be terminated.

2. According to claim 1, the method for deep learning geotechnical stress and strain based on Bi-LSTM fusion CBAM is characterized in that: The Bi-LSTM in S2 uses RNN and its variant LSTM to process data information, which has a bidirectional structure. The basic architecture of the LSTM consists of five parts: input x t 、output h t , Update gate i t 、Forget gate t , output gate o t ,in: Forget Gate t The expression of is as follows: f t =σ(W f *[h t-1 ,x t ]+b f ) Update Gate t The expression of is as follows: i t =σ(W i *[h t-1 ,x t ]+b i ) c' t =tanh(W c *[h t-1 ,x t ]+b c ) c t =f t *c t-1 +i t *c' t Output gate o t The expression of is as follows: the t =σ(W o *[h t-1 ,x t ]+b o ) h t =o t *tanh(c t )。

Citation Information

Cited By

  • Geological disaster prediction method based on transformer and deep learning

    CN119692546A

  • A geological disaster prediction method based on transformer and deep learning

    CN119692546B