Self-adjusting Method for Geometric Parameters of Artificial Bubble Soil Subgrade Structure Based on Deep Learning

Abstract

The present invention discloses a self-adjusting method for geometric parameters of an artificial bubble soil subgrade structure based on deep learning, which relates to the technical field of subgrade structures and includes the following steps: S1, collecting the structural design drawings of the cross-section of the artificial bubble soil road and dividing them into a training set and a test set; S2, constructing an artificial bubble soil data prediction model, processing the training set and the test set to obtain predicted values; S3, obtaining weight values according to the predicted values and using the weight values to adjust the geometric parameters of the cross-section of the artificial bubble soil road. The present invention quickly generates a corresponding cross-section design drawing from the generated text data matrix. Compared with the traditional image generation algorithm, this algorithm greatly improves the data accuracy and accuracy of the generated cross-section drawing and also improves the efficiency of image generation.

Description

Technical Field

[0001] The present invention relates to the technical field of subgrade structures, and particularly relates to a method for self-adjusting geometric parameters of an artificial foamed soil subgrade structure based on deep learning. Background Art

[0002] As the foundation of the road surface, a durable, firm and stable subgrade is the key to ensuring the road surface quality. In mountainous areas with criss-crossing ravines and gullies, bridges or high-fill subgrades are usually used to cross low-lying areas. Due to topographical restrictions such as farmland and ecological forests, when using retaining walls to "reduce the slope" in sections with high subgrades, the filling cannot be compacted due to limited construction space, and the economy is relatively low.

[0003] As a new type of geotechnical material, artificial foamed soil mainly works by fully foaming the foaming agent mechanically through the foaming system of a bubble machine, evenly mixing the foam with cement slurry, and then performing in-situ construction or mold forming through the pumping system of the bubble machine, and forming a new type of lightweight geotechnical material containing a large number of closed air holes after natural curing. It can form closed foam holes inside the concrete, making the concrete lightweight. And its wet density determines its strength. By changing the proportion of various components in the lightweight soil, its strength can be adjusted within a certain range to meet the strength requirements of various subgrade structures. Compared with other materials, it has the advantages of light weight, adjustable density and strength, high fluidity, and convenient construction.

[0004] In recent years, with the booming development of artificial intelligence technology, computer deep learning technology has great potential to empower subgrade structures, thus realizing the automation and intelligence of structural design. The traditional subgrade structure design method is cumbersome, requires a lot of manual intervention, and the subjective awareness of the design institute will affect the parameters of the designed structure, thus affecting the structural performance. Therefore, it is proposed to adopt intelligent structural design by using the determined subgrade structure as a training set to establish a neural network training model. After the model training is completed, the required requirements are input to directly obtain the subgrade design structure. In this way, in the face of actual engineering situations, the on-site design structure can be quickly obtained, greatly reducing the mechanical workload of the design and improving the design efficiency. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a method for self-adjusting geometric parameters of an artificial foamed soil subgrade structure based on deep learning.

[0006] The technical solution of the present invention is: A method for self-adjusting geometric parameters of an artificial foamed soil subgrade structure based on deep learning includes the following steps:

[0007] S1. Collect the structural design drawings of the cross-section of the artificial foamed soil road, and divide them into a training set and a test set;

[0008] S2. Construct an artificial bubble soil data prediction model, process the training set and the test set, and obtain the predicted values;

[0009] S3. Obtain the weight values according to the predicted values, and use the weight values to adjust the geometric parameters of the cross-section of the artificial bubble soil road.

[0010] Further, S1 includes the following sub-steps:

[0011] S11. Collect the structural design drawings of the cross-section of the artificial bubble soil road, and convert the structural design drawings into a text data matrix;

[0012] S12. Perform abnormal data elimination processing and missing data supplementation processing on the text data matrix to complete the preprocessing;

[0013] S13. Divide the preprocessed text data matrix into a training set and a test set.

[0014] Further, in S11, the text data matrix has the following expression:

[0015]

[0016] In the formula, σ(·) represents forming a data matrix from the obtained data, i represents the graph signal number of the structural design drawing of the artificial bubble soil subgrade cross-section input, j represents the number of the artificial bubble soil subgrade cross-section data matrix of the learnable parameters, U represents the eigenvector matrix of the eigenvalue sorting of the artificial bubble soil subgrade cross-section data matrix, represents the artificial bubble soil subgrade cross-section data matrix of the learnable parameters, and T represents the matrix transpose; represents the graph signal of the structural design drawing of the artificial bubble soil subgrade cross-section input, and f k represents the number of graph signal channels of the output structural design drawing of the artificial bubble soil subgrade cross-section, k represents the artificial bubble soil-related data index, and f k-1 represents the number of graph signal channels of the structural design drawing of the artificial bubble soil subgrade cross-section input.

[0017] Further, in S2, the expression of the artificial bubble soil data prediction model y is:

[0018]

[0019] In the formula, M represents the total number of data, λ m represents the first Lagrange multiplier for solving the artificial bubble soil set parameter constraint problem, λ' m represents the second Lagrange multiplier for solving the artificial bubble soil set parameter constraint problem, and x mIt represents the m-th data, x represents the average value of the data, B represents the bias of the discreteness of the artificial bubble soil test, k(·) represents the distribution probability corresponding to the m-th data, exp(·) represents the exponential function, and σ represents the smoothness parameter of the discreteness of the artificial bubble soil test.

[0020] Furthermore, S3 includes the following sub-steps:

[0021] S31. Based on the predicted value, divide the most unfavorable section of the artificial bubble soil road section into several strips and calculate the safety factor of the most unfavorable section;

[0022] S32. Calculate the compressive strength of the subgrade filling body and the self-standing stable compressive strength of the subgrade filling body according to the safety factor of the most unfavorable section;

[0023] S33. Calculate the weight value according to the safety factor of the most unfavorable section, the compressive strength of the subgrade filling body and the self-standing stable compressive strength of the subgrade filling body;

[0024] S34. Adjust the geometric parameters of the artificial bubble soil road section by using the weight value.

[0025] Furthermore, in S31, the safety factor F s of the most unfavorable section is calculated by the formula:

[0026]

[0027] In the formula, m an represents the calculation coefficient of the n-th strip, c' n represents the cohesion of the n-th strip, b n represents the length of the n-th strip, W n represents the weight of the n-th strip, u n represents the pore water pressure of the n-th strip, t represents the tangential component of the strip weight, g represents the acceleration due to gravity, φ' n represents the effective internal friction angle of the n-th strip, α n represents the inclination angle at the bottom of the n-th strip, Q n represents the horizontal acting force of the n-th strip, e n represents the normal acting force of the n-th strip, and R represents the slip surface radius of the most unfavorable section of the artificial bubble soil subgrade.

[0028] Furthermore, in S32, the calculation formula for the compressive strength qu1 of the subgrade filling body is:

[0029]

[0030] In the formula, F S represents the safety factor of the most unfavorable section, and CBR represents the California Bearing Ratio.

[0031] Furthermore, in S32, the calculation formula of the self-supporting compressive strength qu2 of the roadbed filling body is:

[0032] qu2=F S (0.5rH+W);

[0033] In the formula, F S represents the safety factor of the most unfavorable section, r represents the wet bulk density, H represents the filling height, and W represents the load on the top of the filling body.

[0034] Furthermore, in S33, the calculation formula of the weight value K is:

[0035] K=u(a·F s +b·qu1+c·qu2)+v(d·H+e·P+f·U);

[0036] Wherein, a represents the first weight coefficient, b represents the second weight coefficient, c represents the third weight coefficient, d represents the fourth weight coefficient, e represents the fifth weight coefficient, f represents the sixth weight coefficient, u represents the seventh weight coefficient, v represents the eighth weight coefficient, and F s represents the safety factor of the most unfavorable section, qu1 represents the compressive strength of the roadbed fill, qu2 represents the self-supporting stable compressive strength of the roadbed fill, H represents the engineering cost of the adopted scheme, P represents the quality control cost of the adopted scheme, and U represents the time cost consumed by the progress of the adopted scheme.

[0037] The beneficial effects of the present invention are:

[0038] (1) The present invention converts the collected drawing information into a text data matrix, which greatly reduces the storage space of the training samples. At the same time, the model is generated and trained by the text data matrix, which greatly improves the training speed and generation speed of the model compared to image training;

[0039] (2) The present invention uses known complete data matrix training samples to eliminate abnormal data matrices and supplement missing data matrices caused by image recognition errors and partial image loss, which greatly ensures the integrity and reliability of the data;

[0040] (3) The present invention obtains the hyperplane function of artificial bubble soil and automatically optimizes the algorithm. Compared with the common algorithm model, the model greatly improves the convergence and optimization efficiency of the objective function; the present invention compares the economic benefits and strength stability of multiple different design schemes generated by the model, and sets different weight ratios according to requirements. The final weight value can more intuitively reflect the advantages and disadvantages of different design schemes, and ensure the effectiveness and practicality of the design results in actual engineering;

[0041] (4) When the present invention performs calculations, it will automatically adjust each parameter according to the required weight value on-site, without the need to consume a large amount of manpower for repeated trial calculations and iterative modifications, reducing the workload of designers, lowering the design threshold, and improving the design efficiency; the present invention quickly generates a corresponding cross-section design drawing from the generated text data matrix. Compared with the traditional image generation algorithm, this algorithm greatly improves the data accuracy and accuracy of the generated cross-section drawing and also improves the efficiency of image generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a method for self-adjusting geometric parameters of an artificial bubble soil subgrade structure based on deep learning;

[0043] Figure 2 is a schematic diagram of the subgrade cross-section structure when using artificial bubble soil to widen the old subgrade;

[0044] Figure 3 is a schematic diagram of the subgrade cross-section structure for establishing an artificial bubble soil subgrade on a deep soft soil foundation;

[0045] Among them, 1. Road surface, 2. Artificial bubble soil, 3. Old subgrade, 4. Crushed stone leveling layer, 5. Filling soil, 6. Deep soft soil foundation. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] The following further describes the embodiments of the present invention with reference to the accompanying drawings.

[0047] As Figure 1 shown, the present invention provides a method for self-adjusting geometric parameters of an artificial bubble soil subgrade structure based on deep learning, including the following steps:

[0048] S1. Collect the structural design drawings of the artificial bubble soil road section and divide them into a training set and a test set;

[0049] S2. Construct an artificial bubble soil data prediction model, process the training set and the test set to obtain predicted values;

[0050] S3. Obtain weight values according to the predicted values and use the weight values to adjust the geometric parameters of the artificial bubble soil road section.

[0051] In the embodiment of the present invention, S1 includes the following sub-steps:

[0052] S11. Collect the structural design drawings of the artificial bubble soil road section and convert the structural design drawings into a text data matrix;

[0053] S12. Perform abnormal data elimination processing and missing data supplementation processing on the text data matrix to complete preprocessing;

[0054] S13. Divide the preprocessed text data matrix into a training set and a test set.

[0055] The present invention collects the subgrade cross-section information of the currently adopted artificial bubble soil. Since it is considered that in actual engineering, on-site construction is mainly carried out through design drawings, when collecting the cross-section information of the artificial bubble soil, the collected subgrade cross-section structure design drawing is used as a training sample, and it is classified into different scenarios and conditions. Then, through the GN algorithm, the design drawing training sample is converted into a text data matrix, and the text data matrix is collected and stored. The text data matrix should include the on-site situation of the design drawing, the length and thickness design data of the gravel leveling layer at the bottom of the subgrade, the length and thickness design data of the artificial bubble soil, the inclination angle and width design data of the external fill soil, etc., which is convenient for subsequent training and rapid processing of the model.

[0056] After calculation by the above algorithm, a text data matrix generated from the subgrade cross-section structure design drawing of the artificial bubble soil is obtained, and this data matrix is collected and stored as a training sample. Then, for the collected data matrix, the abnormal data matrix caused by image recognition errors and the missing part of the data matrix caused by partial image loss are excluded and supplemented with the known complete data matrix training sample.

[0057] Then, a correlation analysis is performed on each artificial bubble soil subgrade cross-section data parameter in the data matrix to ensure that there is no obvious multicollinearity between the final output variables. When the correlation coefficient value between two artificial bubble soil variables is less than 0.7, it indicates that there is no obvious correlation between these two variables. The calculation formula is as follows:

[0058]

[0059] In the formula, R represents the correlation coefficient between two artificial bubble soil variables, x1 and x2 represent two variables, Cov(x1, x2) represents the covariance between the two artificial bubble soil variables x1 and x2, E(x1) represents the expectation of x1, E(x2) represents the expectation of x2, S 2 (x i ) represents the variance of the artificial bubble soil variable x i S 2 (x j ) represents the variance of the artificial bubble soil variable x j variable.

[0060] Each data matrix in the processed data set matrix contains multiple input variable features, namely: the actual on-site situation, the subgrade structure dimensions, the load information, and the boundary conditions. The output results are the length and thickness of the gravel leveling layer at the bottom of the subgrade, the length and thickness of the artificial aerated soil, the length and width of the stepped structure of the artificial aerated soil, the inclination angle of the external fill, and the width of the external fill, etc. These data will be divided into a training set and a test set according to a ratio of 7:3.

[0061] In the embodiment of the present invention, in S11, the text data matrix has the following expression:

[0062]

[0063] In the formula, σ(·) represents forming a data matrix from the obtained data. i represents the graph signal number of the input cross-section structure design drawing of the artificial aerated soil subgrade. j represents the number of the data matrix of the artificial aerated soil subgrade cross-section with learnable parameters. U represents the eigenvector matrix of the eigenvalue sorting of the data matrix of the artificial aerated soil subgrade cross-section. represents the data matrix of the artificial aerated soil subgrade cross-section with learnable parameters, and T represents matrix transpose; represents the graph signal of the input cross-section structure design drawing of the artificial aerated soil subgrade, and f k represents the number of channels of the graph signal of the output cross-section structure design drawing of the artificial aerated soil subgrade. k represents the index of the data related to the artificial aerated soil, and f k-1 represents the number of channels of the graph signal of the input cross-section structure design drawing of the artificial aerated soil subgrade.

[0064] In the embodiment of the present invention, in S2, for the sorted training set and test set, while obtaining the hyperplane function of the artificial aerated soil through the ML-LM algorithm, the algorithm is automatically optimized. An artificial aerated soil data prediction model is established, and according to the weights of the various data of the artificial aerated soil, it is transformed into a dual problem to determine the predicted value. The expression of the artificial aerated soil data prediction model y is:

[0065]

[0066] In the formula, M represents the total number of data, and λ m represents the first Lagrange multiplier for solving the artificial aerated soil set parameter constraint problem, and λ' m represents the second Lagrange multiplier for solving the artificial aerated soil set parameter constraint problem. x m represents the m-th data, x represents the average value of the data, B represents the bias of the discrete degree of the artificial aerated soil test, k(·) represents the distribution probability corresponding to the m-th data, exp(·) represents the exponential function, and σ represents the smoothing degree parameter of the discrete degree of the artificial aerated soil test.

[0067] The coefficient of determination R 2 The accuracy of the prediction model fitted by the ML-LM algorithm is evaluated by using the root mean square error (RMSE). The calculation formula is as follows:

[0068]

[0069] Among them, y i represents the true value of the ith data in the collected artificial aerated soil data, y' i represents the predicted value of the ith data after solving the problem of collective parameter constraints of artificial bubble soil, Represents the true average value of all the data collected for artificial aerated soil, R 2 The closer it is to 1, the better the model fits the data and the higher the prediction accuracy. RMSE expresses the error between the predicted value and the true value of the model. The smaller the value, the smaller the deviation and the higher the accuracy of the prediction model.

[0070] Subsequently, based on the predicted values ​​of the above-mentioned prediction model, a deep learning network model of artificial bubble soil was established, and mechanical verification was performed on the artificial bubble soil road section generated by the deep learning network model. The mechanical verification included stability verification and strength verification to ensure the effectiveness of the design scheme of intelligent artificial bubble soil.

[0071] In this embodiment of the present invention, S3 includes the following sub-steps:

[0072] S31, based on the predicted value, dividing the most unfavorable section of the artificial bubble soil road section into a number of strips and blocks, and calculating the safety factor of the most unfavorable section;

[0073] S32. Calculate the compressive strength of the roadbed fill and the self-supporting compressive strength of the roadbed fill based on the safety factor of the most unfavorable section;

[0074] S33. Calculate the weight value according to the safety factor of the most unfavorable section, the compressive strength of the roadbed fill and the self-supporting compressive strength of the roadbed fill;

[0075] S34. Use weight values ​​to adjust geometric parameters of the artificial bubble soil road section.

[0076] In the embodiment of the present invention, in S31, the safety factor F of the most unfavorable section s The calculation formula is:

[0077]

[0078] In the formula, m an Indicates the calculation coefficient of the nth bar, c' n represents the cohesion of the nth bar, b nrepresents the length of the nth strip, W n represents the weight of the nth strip, u n represents the pore water pressure of the nth strip, t represents the tangential component of the strip weight, g represents the acceleration due to gravity, φ' n represents the effective internal friction angle of the nth strip, α n represents the inclination angle at the bottom of the nth strip, Q n represents the horizontal acting force of the nth strip, e n represents the normal acting force of the nth strip, R represents the slip surface radius of the most unfavorable section of the artificial bubble soil subgrade.

[0079] In the embodiment of the present invention, in S32, the calculation formula for the compressive strength qu1 of the subgrade filling body is:

[0080]

[0081] In the formula, F S represents the safety factor of the most unfavorable section, and CBR represents the California Bearing Ratio.

[0082] In the embodiment of the present invention, in S32, the calculation formula for the self-standing stable compressive strength qu2 of the subgrade filling body is:

[0083] qu2 = F S (0.5rH + W);

[0084] In the formula, F S represents the safety factor of the most unfavorable section, r represents the wet unit weight, H represents the filling height, and W represents the load at the top of the filling body.

[0085] In the embodiment of the present invention, in S33, for conservative calculation, the strength and stability of the most unfavorable section of the generated artificial bubble soil subgrade need to be 40% - 50% higher than the strength and stability of the specification to ensure the effectiveness of the design scheme; secondly, the model compares the strength stability and economic benefits of the generated multiple different design schemes, conducts a comparative analysis on aspects such as project cost, quality control, and progress control in terms of economic benefits, and sets different weight ratios according to requirements. Finally, a weight value is obtained, and the calculation formula for the weight value K is:

[0086] K = u(a·F s + b·qu1 + c·qu2)+ v(d·H + e·P + f·U);

[0087] In the formula, a represents the first weight coefficient, b represents the second weight coefficient, c represents the third weight coefficient, d represents the fourth weight coefficient, e represents the fifth weight coefficient, f represents the sixth weight coefficient, u represents the seventh weight coefficient, v represents the eighth weight coefficient, F sIt represents the safety factor of the most unfavorable cross-section, qu1 represents the compressive strength of the subgrade filling body, qu2 represents the self-stabilizing compressive strength of the subgrade filling body, H represents the project cost of the adopted plan, P represents the quality control cost of the adopted plan, and U represents the time cost consumed by the progress of the adopted plan.

[0088] Finally, according to different requirements on-site, a weight value within a certain range can be taken. When this model calculates, it will automatically adjust each parameter according to the required weight value on-site. The formula for its automatic adjustment is:

[0089] if(K<K 下 );

[0090]

[0091] if(K>K 上 );

[0092]

[0093] Among them, K 上 and K 下 respectively represent the upper and lower limits of the weight value within the required weight range. ΔF s , Δqu1, Δqu2, ΔH, ΔP, ΔU respectively represent the magnitudes of the self-adjusted F s , qu1, qu2, H, P, U.

[0094] It should be noted that when the weight value of the plan calculated by the model is lower than the lower limit, the parameters will be adjusted at this time. First, the safety factor, the compressive strength of the filling body, and the self-stabilizing compressive strength of the filling body will be increased, and the project cost, quality control cost, and time cost will be adjusted according to the required weight value; when the weight value of the plan calculated by the model is higher than the upper limit, first, the project cost, quality control cost, and time cost will be reduced, and a small adjustment will be made to them on the premise of ensuring that the safety factor, the compressive strength of the filling body, and the self-stabilizing compressive strength of the filling body meet the requirements. The parameters will be automatically adjusted iteratively, and the step size of each iteration is about one-tenth of the initial parameter. Then, the magnitude of K will be calculated again and compared again until the constraint conditions are met, and the plan that meets the conditions will be selected on-site;

[0095] Subsequently, through the automatic adjustment of the cross-section geometric parameters by the above algorithm, the geometric structure values of multiple cross-sections with weight values within the required range are obtained, including the length L1 and thickness d1 of the gravel leveling layer at the bottom of the subgrade, the length L2 and thickness d2 of the artificial aerated soil, the length L3 and width d3 of the stepped structure of the artificial aerated soil, the inclination 1:m of the external fill soil, the width H1 of the external fill soil, etc. According to the generated subgrade cross-section set structure data and the known on-site data, the text data matrix is generated into the corresponding on-site cross-section design image through the IG-SD algorithm, which is convenient for the on-site personnel to understand.

[0096] The finally generated on-site cross-section structure design drawing of the artificial aerated soil subgrade is as Figure 2 and Figure 3 shown. The figure respectively exemplarily shows the schematic diagram of the subgrade cross-section structure when the artificial aerated soil is used to widen the old subgrade and the schematic diagram of the subgrade cross-section structure of the artificial aerated soil subgrade built on the deep soft soil foundation. In the figure, 1 is the road surface of the artificial aerated soil subgrade, 2 is the artificial aerated soil poured in the artificial aerated soil subgrade, 3 is the on-site old subgrade, 4 is the gravel leveling layer required for the artificial aerated soil subgrade, 5 is the fill soil required for the artificial aerated soil subgrade, and 6 is the deep soft soil foundation.

[0097] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on these technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. A method for self-adjusting geometric parameters of artificial bubble soil roadbed structure based on deep learning, characterized in that: The following steps are involved: S1, collect the structural design drawings of the artificial bubble soil road section and divide them into a training set and a test set; S2, constructing an artificial bubble soil data prediction model, processing the training set and the test set, and obtaining the predicted value; S3, obtaining a weight value according to the predicted value, and adjusting the geometric parameters of the cross section of the artificial bubble soil road using the weight value; In S2, artificial aerated soil data prediction model The expression is: ； ； In the formula, Indicates the total number of data. represents the first Lagrange multiplier for solving the problem of set parameter constraints on artificial bubble soil, represents the second Lagrange multiplier for solving the problem of set parameter constraints on artificial bubble soil, Indicates data, represents the average value of the data, The bias that represents the discrete degree of the artificial bubble soil test, Indicates The distribution probability corresponding to the data is represents the exponential function, Smoothness parameter indicating the discreteness of artificial bubble soil test; The S3 comprises the following sub-steps: S31, based on the predicted value, dividing the most unfavorable section of the artificial bubble soil road section into a number of strips and blocks, and calculating the safety factor of the most unfavorable section; S32. Calculate the compressive strength of the roadbed fill and the self-supporting compressive strength of the roadbed fill based on the safety factor of the most unfavorable section; S33. Calculate the weight value according to the safety factor of the most unfavorable section, the compressive strength of the roadbed fill and the self-supporting compressive strength of the roadbed fill; S34, adjusting the geometric parameters of the artificial bubble soil road section using the weight value; In S31, the safety factor of the most unfavorable section The calculation formula is: ； In the formula, Indicates The calculation coefficient of each bar, Indicates The cohesion of the strips, Indicates The length of the bar, Indicates The weight of the block, Indicates The pore water pressure of each strip, represents the tangential component of the bar weight, represents the acceleration due to gravity, Indicates The effective internal friction angle of a bar is Indicates The inclination angle of the bottom of the bar, Indicates The horizontal force acting on the bar, Indicates The normal force on a bar, Indicates the sliding surface radius of the most unfavorable section of the artificial bubble soil roadbed; In S32, the compressive strength of the roadbed fill The calculation formula is: ； In the formula, The safety factor of the most unfavorable section, It represents the California carrying ratio; In S32, the self-stable compressive strength of the roadbed fill body The calculation formula is: ； In the formula, The safety factor of the most unfavorable section, It represents the wet bulk density. Indicates the filling height, represents the load on the top of the fill; In S33, the weight value The calculation formula is: ； In the formula, represents the first weight coefficient, represents the second weight coefficient, represents the third weight coefficient, represents the fourth weight coefficient, represents the fifth weight coefficient, represents the sixth weight coefficient, represents the seventh weight coefficient, represents the eighth weight coefficient, The safety factor of the most unfavorable section, Indicates the compressive strength of roadbed fill. Indicates the self-supporting compressive strength of the roadbed fill. Indicates the engineering cost of the adopted solution, represents the quality control cost of the adopted solution, Indicates the time cost of the progress of the adopted solution.

2. The method for self-adjusting geometric parameters of artificial bubble soil roadbed structure based on deep learning according to claim 1 is characterized in that: The S1 comprises the following sub-steps: S11, collecting structural design drawings of the artificial bubble soil road section, and converting the structural design drawings into a text data matrix; S12, performing abnormal data elimination processing and missing data supplement processing on the text data matrix to complete preprocessing; S13. Divide the preprocessed text data matrix into a training set and a test set.

3. The method for self-adjusting geometric parameters of artificial bubble soil roadbed structure based on deep learning according to claim 2 is characterized in that: In S11, the text data matrix The expression is: ； In the formula, Indicates that the obtained data is formed into a data matrix, Indicates the drawing signal number of the input artificial bubble soil roadbed section structure design drawing. The number of the artificial bubble soil roadbed section data matrix representing the learnable parameters, The eigenvector matrix representing the eigenvalue sorting of the artificial bubble soil roadbed cross-section data matrix, The artificial bubble soil roadbed section data matrix representing the learnable parameters, Represents matrix transpose; The graphic signal representing the input of the cross-section structural design drawing of the artificial bubble soil roadbed. Indicates the number of image signal channels for outputting the cross-section structure design drawings of artificial bubble soil roadbed. Indicates the data index related to artificial bubble soil. Indicates the number of drawing signal channels for inputting the design drawing of the cross-section structure of the artificial bubble soil roadbed.

Images