Coarse-grained soil permeability coefficient intelligent prediction method based on multi-data fusion

Abstract

The invention relates to the field of coarse-grained soil permeability coefficient prediction, in particular to an intelligent coarse-grained soil permeability coefficient prediction method based on multi-data fusion, and the method comprises the following steps: calculating actually measured void ratio data based on dielectric constant data, calculating actually measured fine particle content data based on conductivity data, and calculating actually measured particle size based on scattered light intensity distribution data; a first grading medium coarse-grained soil permeability coefficient is generated by improving a Hazen model, and the improved Hazen model comprises a dynamic fine-grained blockage coefficient and a dynamic fine-grained contact coefficient; a second grading medium coarse-grained soil permeability coefficient is generated by improving a USBR model, and the improved USBR model comprises grading parameters; and generating the permeability coefficient of the medium coarse-grained soil in prediction grading through a weighted model based on a convolutional neural network architecture. According to the method, the empirical model is combined with the intelligent algorithm, so that the problems of low test efficiency and poor dynamic adaptability of the empirical model to the graded medium soil are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of permeability coefficient prediction for coarse-grained soil, and more particularly to an intelligent prediction method for coarse-grained soil permeability coefficient based on multi-data fusion. Background Technology

[0002] In the construction of numerous geotechnical engineering projects, such as roadbed engineering, dam seepage prevention, foundation pit excavation, and groundwater resource development, coarse-grained soil is widely used as fill material, dam shell material, and drainage material due to its excellent engineering properties, including high strength, low compressibility, and good permeability. The permeability coefficient of coarse-grained soil, as a core hydraulic parameter characterizing its permeability, directly determines the seepage prevention design scheme, seepage stability assessment results, and construction quality control standards of the project. For example, in highway roadbed construction, if the permeability coefficient of coarse-grained soil is too high, it can easily lead to rainwater infiltration, resulting in a decrease in roadbed bearing capacity and uneven settlement. In reservoir dam engineering, accurate control of the permeability coefficient of the coarse-grained soil in the dam shell is a crucial prerequisite for preventing dam leakage and avoiding seepage damage.

[0003] Therefore, achieving efficient and accurate prediction of the permeability coefficient of coarse-grained soil is crucial for ensuring the construction safety of geotechnical engineering projects.

[0004] Currently, the methods for obtaining the permeability coefficient of coarse-grained soil in the engineering field are mainly divided into classical empirical model prediction methods and prediction methods based on intelligent algorithms. However, both types of methods are difficult to meet the real-time and high-precision requirements of modern engineering for the prediction of the permeability coefficient of coarse-grained soil with medium gradation.

[0005] In the classification standards of civil engineering, medium-graded soil is a transitional type of sand containing fine particles within coarse-grained soil (particles with a diameter greater than 0.075 mm account for more than 50% of the total mass). The definition of medium-graded soil is: sand particles (0.075 to 2 mm in diameter) with a mass fraction of 50% to 85%, coarse particles with a mass fraction greater than 50%, and fine particles with a mass fraction of 5% to 30%. It is evident that medium-graded soil possesses a unique transitional structure with a coarse-grained skeleton and a partially filled fine-grained portion. The residual large pores between coarse particles provide the primary permeability channels, while the tiny pores between fine particles provide secondary, high-resistance permeability channels. Therefore, its coarse-grained soil permeability coefficient differs significantly from both poorly graded homogeneous sand (dominated by large pores) and well-graded soil (completely filled with fine particles), exhibiting unique transitional permeability characteristics. Consequently, real-time and accurate measurements during construction cannot be simply performed using classic empirical models and conventional intelligent algorithms applicable to pure sand or well-graded soil.

[0006] Specifically, classical empirical model prediction methods are based on classical empirical models of the physical and mechanical parameters of coarse-grained soils. Among these, the Hazen model and the USBR model are the two most widely used prediction models. The Hazen model is based on the correlation between effective particle size and permeability coefficient. It is simple in form and convenient to calculate, but its core drawback is that the model parameters are fixed values. It does not consider dynamic factors such as the pore blockage effect caused by fine particle inhomogeneity during engineering construction and changes in the contact state between fine particles, resulting in low accuracy in predicting the permeability coefficient of medium-graded coarse-grained soils. The traditional USBR model uses a static calculation method and cannot adapt to the dynamic fluctuations in the particle size distribution of coarse-grained soils during construction. Especially under the condition that the fine particle content of medium-graded soils is unstable, the prediction error increases significantly.

[0007] In recent years, with the development of artificial intelligence technology, methods for predicting the permeability coefficient of coarse-grained soil based on intelligent algorithms have gradually emerged. These methods improve prediction accuracy by mining the nonlinear mapping relationship between multi-source data. However, the integration between existing intelligent models and classical empirical models is insufficient. Existing methods are purely data-driven black-box models that do not fully utilize the physical mechanisms of classical empirical models. This results in a lack of physical constraints in the prediction results of coarse-grained soil permeability coefficient, which may violate the basic laws of geotechnical mechanics. Furthermore, the model's generalization ability is limited, making it difficult to adapt to the permeability coefficient prediction needs of coarse-grained soil in different construction areas and with different gradations.

[0008] Therefore, how to combine empirical model prediction methods with intelligent algorithms to avoid the low efficiency, poor dynamic adaptability, and insufficient prediction accuracy of empirical models in multi-data experimental measurement is a technical problem that needs to be solved. Summary of the Invention

[0009] To address this, the present invention provides an intelligent prediction method for the permeability coefficient of coarse-grained soil based on multi-data fusion. By accurately acquiring multi-data from multiple sources, dynamically predicting using improved Hazen and USBR models, and optimizing using weighted fusion of convolutional neural networks, the empirical model is combined with intelligent algorithms, effectively solving the problems of low experimental efficiency, poor dynamic adaptability, and insufficient prediction accuracy of the empirical model for medium-graded soils.

[0010] To achieve the above objectives, this invention proposes an intelligent prediction method for the permeability coefficient of coarse-grained soil based on multi-data fusion, comprising: The measured porosity data is calculated based on the dielectric constant data collected by the TDR sensor in the construction area, the measured fine particle content data is calculated based on the conductivity data collected by the conductivity sensor in the construction area, and the measured particle size is calculated based on the scattered light intensity distribution data collected by the laser diffraction sensor in the construction area. The measured void ratio data, measured fine particle content data, and measured particle size are used to generate the permeability coefficient of medium-coarse-grained soil in the first grade by using an improved Hazen model, wherein the improved Hazen model includes dynamic fine particle blockage coefficient and dynamic fine particle contact coefficient. The measured void ratio data, measured fine particle content data, and measured particle size are used to generate the permeability coefficient of the second-grade medium-coarse-grained soil by using an improved USBR model, wherein the improved USBR model includes gradation parameters. The permeability coefficients of the first and second graded medium-coarse-grained soils are used to generate the predicted permeability coefficients of the graded medium-coarse-grained soils through a weighted model based on a convolutional neural network architecture.

[0011] Furthermore, the process of generating the permeability coefficient of the first-grade medium-coarse-grained soil includes: The fine particle activity coefficient is calculated based on the measured porosity data and the measured fine particle content data, and the dynamic fine particle blockage coefficient is calculated based on the measured fine particle content data and the measured porosity data. The calibration permeability coefficient of medium-strength soil was determined by conducting permeability tests based on Darcy's law with temperature correction in the construction area. Based on the measured fine particle content data, measured particle size, fine particle activity coefficient, measured permeability coefficient of medium soil and dynamic fine particle blockage coefficient, the dynamic fine particle content coefficient is generated by least squares fitting method. The measured fine particle content data, measured particle size, fine particle activity coefficient, dynamic fine particle blockage coefficient, and dynamic fine particle content coefficient are substituted into the improved Hazen model to generate the permeability coefficient of the medium-coarse-grained soil in the first grade.

[0012] Furthermore, the process of substituting the improved Hazen model into the process of generating the permeability coefficient of the first-graded medium-coarse-grained soil includes: The improved Hazen model calculates the negative exponent value of the product of the dynamic fine-particle blockage coefficient, the measured fine-particle content data, and the fine-particle activity coefficient, calculates the product of the dynamic fine-particle content coefficient and the square of the measured fine-particle content data, and generates the Hazen correction coefficient for graded medium soil based on the ratio of the negative exponent value to the product. The improved Hazen model calculates an index value based on measured fine particle content data and a blockage nonlinearity index. The fine particle content blockage term is generated by multiplying the index value and a blockage degree parameter, wherein the blockage degree parameter is determined by settlement tests in the construction area. The improved Hazen model is based on the product of contact effect parameters, fine particle activity coefficient, and measured particle size to generate the basic permeability term; The improved Hazen model is based on the product of the Hazen correction coefficient for medium-graded soil, the fine-grained content blocking term, and the basic permeability term to generate the permeability coefficient for medium-coarse-grained soil of the first grade.

[0013] Furthermore, the process of calculating the fine-grained activity coefficient includes: A first ratio is calculated between the current fine-grained content data and the maximum value of historical fine-grained content data. Based on the average of the maximum and minimum values ​​of historical void ratio data, a coarse-grained soil reference void ratio is generated. A second ratio is calculated between the coarse-grained soil reference void ratio and the current void ratio data. Based on the product of the first and second ratios, the fine-grained activity coefficient is determined. The measured fine-grained content data includes the current fine-grained content data and the maximum value of historical fine-grained content data. The measured void ratio data includes the current void ratio data and historical void ratio data. The process of calculating the dynamic fine-particle blockage coefficient includes: The ratio of the controlled particle size to the effective particle size and the current porosity data are substituted into the attenuation coefficient calibration formula to generate a dynamic fine particle blockage coefficient, wherein the current porosity data includes the controlled particle size and the effective particle size.

[0014] Furthermore, the process for determining the calibration permeability coefficient of moderate soil includes: The initial intermediate soil calibration permeability coefficient is obtained from Darcy's law of the permeability test. The third ratio of the current temperature hydrodynamic viscosity to the standard temperature hydrodynamic viscosity is calculated. Based on the product of the initial intermediate soil calibration permeability coefficient and the third ratio, the intermediate soil calibration permeability coefficient is generated. The process of generating the dynamic fine-grain content coefficient using the least squares fitting formula includes: The dependent variable is constructed based on the effective particle size, the measured permeability coefficient of medium soil, the dynamic fine-particle blockage coefficient, the measured void ratio data, and the fine-particle activity coefficient. The independent variable is constructed based on the square of the measured fine-particle content data. The fitting formula is constructed based on the fact that the dependent variable is equal to the product of the independent variable and the dynamic fine-particle content coefficient. The fitting formula is solved by the weighted least squares method of minimizing the weighted sum of squared residuals to generate the dynamic fine-particle content coefficient.

[0015] Furthermore, the process of generating the permeability coefficient of the second-grade medium-coarse-grained soil includes: The improved USBR model calculates the gradation particle size parameters based on the controlled particle size and effective particle size, calculates the gradation void ratio based on the measured void ratio data and the measured fine particle content data, and generates the permeability coefficient of the second gradation medium-coarse-grained soil based on the gradation particle size parameters and the gradation void ratio. The gradation parameters include gradation particle size parameters and gradation void ratio.

[0016] Furthermore, the process of generating the predicted permeability coefficient of medium-coarse-grained soil with appropriate gradation includes: Based on the measured void ratio data, measured fine particle content data, measured particle size, permeability coefficient of medium-coarse soil in the first grade and permeability coefficient of medium-coarse soil in the second grade, a measured input list is constructed. The measured input list is passed through multi-scale convolutional units to generate comprehensive features; The comprehensive features are passed through the output mapping layer to generate weighting coefficients, and the permeability coefficients of the first grade medium-coarse-grained soil and the second grade medium-coarse-grained soil are weighted and calculated using the weighting coefficients to generate the predicted permeability coefficient of the grade medium-coarse-grained soil. The weighted model mentioned above includes multi-scale convolutional units and output mapping layers.

[0017] Furthermore, the process of generating comprehensive features includes: The measured input list is passed through a fine-grained scale convolutional layer to generate fine-grained effect features; The measured input list is passed through a gradation scale convolutional layer to generate gradation effect features; The measured input list is passed through a macro-scale convolutional layer to generate comprehensive effect features; The fine-grained effect features, gradation effect features, and combined effect features are vector-concatenated to generate the combined feature; The multi-scale convolutional unit includes fine-grained convolutional layers, graded-scale convolutional layers, and macro-scale convolutional layers.

[0018] Furthermore, the process of generating the predicted permeability coefficient of medium-coarse-grained soil with appropriate gradation also includes: A weight regularization term is constructed based on the weighting coefficients, and a total loss function is constructed based on the weight regularization term and the mean squared error term. The weighted model is then trained based on the total loss function.

[0019] Furthermore, the process of calculating the measured fine particle content data includes: Porosity is calculated based on the conductivity data, the dielectric constant of solid particles, and the dielectric constant of pore water, and the measured fine particle content data is calculated based on the porosity.

[0020] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention avoids the problems of traditional coarse-grained soil permeability coefficient prediction methods relying on a single data source, poor model adaptability, and insufficient prediction accuracy by using multi-source sensor data fusion, improved model optimization, and intelligent weighted fusion. By introducing dynamic fine-grained clogging coefficient and dynamic fine-grained contact coefficient to optimize the Hazen model, and by introducing gradation parameters to optimize the USBR model, both types of models can accurately match the permeability characteristics of medium-graded coarse-grained soils. By using convolutional neural networks to extract core features and perform adaptive weighted fusion of the results from the two models, the empirical model combined with intelligent algorithms effectively solves the problems of low experimental efficiency, poor dynamic adaptability, and insufficient prediction accuracy of empirical models for medium-graded soils.

[0021] In particular, this invention, through an improved Hazen model, comprehensively covers fine-grained clogging, gradation characteristics, and contact effects. The Hazen correction coefficient for medium-graded soil is derived by using the ratio of the negative exponential calculation of the dynamic fine-grained clogging coefficient, fine-grained content, and fine-grained activity coefficient to the square term of fine-grained content, accurately quantifying the correction effect of gradation and the dynamic interaction of fine particles on the permeability coefficient. The fine-grained content clogging term, combined with the clogging degree parameter determined by the settlement test in the construction area, achieves nonlinear quantification of the clogging effect, closely conforming to the complex evolution law of fine-grained clogging in actual engineering. The basic permeability term, through the product calculation of contact effect parameters, fine-grained activity coefficient, and measured particle size, fully captures the fundamental influence of particle contact state on permeability, enabling the improved Hazen model to comprehensively adapt to the permeability characteristics of medium-graded coarse-grained soil. The improved USBR model focuses on the core coupling relationship between gradation and porosity, achieving accurate adaptation to permeability characteristics under different gradation states by controlling the gradation particle size parameter calculated from the particle size and effective particle size, as well as the gradation porosity ratio combined with the measured porosity and fine-grained content.

[0022] In particular, this invention achieves complementary advantages in prediction results by combining the improved Hazen model and the improved USBR model with a subsequent refined intelligent weighted fusion strategy. Through three convolutional layers of multi-scale convolutional units at the fine-grained, gradation, and macro scales, the core features of fine-grained dynamic effects, gradation characteristic effects, and comprehensive geological condition effects are accurately extracted, avoiding the one-sidedness of single-scale feature extraction. This enables adaptive weighted fusion of the results from the two models, meeting the requirements for experimental efficiency and prediction accuracy of medium-graded soils during construction. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the intelligent prediction method for coarse-grained soil permeability coefficient based on multi-data fusion, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the improved Hazen model of the intelligent prediction method for coarse-grained soil permeability coefficient based on multi-data fusion, according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the least squares fitting method of the intelligent prediction method for coarse-grained soil permeability coefficient based on multi-data fusion, according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating the weighted model of the intelligent prediction method for coarse-grained soil permeability coefficient based on multi-data fusion, according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0025] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0026] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0027] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0028] like Figures 1 to 4 As shown, this invention provides an intelligent prediction method for the permeability coefficient of coarse-grained soil based on multi-data fusion. By accurately acquiring multi-data from multiple sources of sensors, dynamically predicting using improved Hazen and USBR models, and optimizing using weighted fusion of convolutional neural networks, the empirical model is combined with intelligent algorithms, effectively solving the problems of low experimental efficiency, poor dynamic adaptability, and insufficient prediction accuracy of the empirical model for medium-graded soils.

[0029] like Figure 1 As shown in the figure, this embodiment proposes an intelligent prediction method for the permeability coefficient of coarse-grained soil based on multi-data fusion, including: The measured porosity data is calculated based on the dielectric constant data collected by the TDR sensor in the construction area, the measured fine particle content data is calculated based on the conductivity data collected by the conductivity sensor in the construction area, and the measured particle size is calculated based on the scattered light intensity distribution data collected by the laser diffraction sensor in the construction area. The measured void ratio data, measured fine particle content data, and measured particle size are used to generate the permeability coefficient of medium-coarse-grained soil in the first grade by using an improved Hazen model, wherein the improved Hazen model includes a dynamic fine particle blockage coefficient and a dynamic fine particle content coefficient. The measured void ratio data, measured fine particle content data, and measured particle size are used to generate the permeability coefficient of the second-grade medium-coarse-grained soil by using an improved USBR model, wherein the improved USBR model includes gradation parameters. The permeability coefficients of the first and second graded medium-coarse-grained soils are used to generate the predicted permeability coefficients of the graded medium-coarse-grained soils through a weighted model based on a convolutional neural network architecture.

[0030] Specifically, the medium-graded soil refers to a transitional sandy soil containing fine particles within a coarse-grained soil (particles with a diameter greater than 0.075 mm account for more than 50% of the total mass). Its sand particle (0.075 to 2 mm) mass fraction is 50% to 85%, coarse particle mass fraction is greater than 50%, and fine particle mass fraction is 5% to 30%. Therefore, this embodiment addresses the characteristics of the variable coarse and fine particle content and the large range and fluctuation of the permeability coefficient of medium-graded coarse-grained soil, which reflects water flow permeability. It uses an improved Hazen model and an improved USBR model to reflect its physical properties, and dynamically weights the data using a weighted model based on a convolutional neural network architecture. This allows the improved Hazen model to quantify the fine-particle blockage effect from the perspectives of microscopic blockage, surface adsorption, and pore blockage, while the improved USBR model quantifies the gradation morphology effect from the perspective of equivalent particle size. The weighted model achieves complementary verification of the physical mechanisms of the improved Hazen and USBR models, thereby making the prediction of the permeability coefficient of medium-graded coarse-grained soil more accurate.

[0031] like Figure 2 As shown, the process of generating the permeability coefficient of the first-grade medium-coarse-grained soil further includes: The fine particle activity coefficient is calculated based on the measured porosity data and the measured fine particle content data, and the dynamic fine particle blockage coefficient is calculated based on the measured fine particle content data and the measured porosity data. The calibration permeability coefficient of medium-strength soil was determined by conducting permeability tests based on Darcy's law with temperature correction in the construction area. Based on the measured fine particle content data, measured particle size, fine particle activity coefficient, measured permeability coefficient of medium soil and dynamic fine particle blockage coefficient, the dynamic fine particle content coefficient is generated by least squares fitting method. The measured fine particle content data, measured particle size, fine particle activity coefficient, dynamic fine particle blockage coefficient, and dynamic fine particle content coefficient are substituted into the improved Hazen model to generate the permeability coefficient of the medium-coarse-grained soil in the first grade.

[0032] Furthermore, the process of substituting the improved Hazen model into the process of generating the permeability coefficient of the first-graded medium-coarse-grained soil includes: The improved Hazen model calculates the negative exponent value of the product of the dynamic fine-particle blockage coefficient, the measured fine-particle content data, and the fine-particle activity coefficient, calculates the product of the dynamic fine-particle content coefficient and the square of the measured fine-particle content data, and generates the Hazen correction coefficient for graded medium soil based on the ratio of the negative exponent value to the product. The improved Hazen model calculates an index value based on measured fine particle content data and a blockage nonlinearity index. The fine particle content blockage term is generated by multiplying the index value and a blockage degree parameter, wherein the blockage degree parameter is determined by settlement tests in the construction area. The improved Hazen model is based on the product of contact effect parameters, fine particle activity coefficient, and measured particle size to generate the basic permeability term; The improved Hazen model is based on the product of the Hazen correction coefficient for medium-graded soil, the fine-grained content blocking term, and the basic permeability term to generate the permeability coefficient for medium-coarse-grained soil of the first grade.

[0033] Specifically, the improved Hazen model can be expressed as: In the formula, This represents the Hazen correction factor for medium-graded soil. The value represents the permeability coefficient of the first-grade medium-coarse-grained soil, and C represents the original Hazen coefficient, preferably 0.01 for clean sand. Indicates the dynamic fine-particle clogging coefficient. This represents the measured fine particle content data. Indicates the fine-particle activity coefficient. These represent the congestion severity parameter and the congestion nonlinearity index, respectively, where the congestion severity parameter is obtained by using... The settlement test of the construction area is used to determine the formula, where Let represent the particle density and the density of water, respectively, preferably 2600 kg / m³ and 1000 kg / m³, and g represent the gravitational acceleration of 9.8 m / s². This represents the square of the typical value of the average fine particle size, which is preferably 0.03 mm. The dynamic viscosity of water is preferably expressed as... , The seepage velocity obtained from the settlement test is represented by the clogging nonlinearity index, which is based on an empirical formula. calculate, The contact effect parameter is preferably 0.1. All are measured particle sizes, among which This indicates the effective particle size (i.e., the particle size smaller than 10% of the mass fraction of particles in historical measured particle size data). This represents the median particle size (i.e., the particle size smaller than 50% of the mass fraction of particles in historical measured particle size data). This represents the dynamic fine particle content coefficient.

[0034] It is understandable that in the improved Hazen model, the Hazen correction coefficient for gradation-moderate soil is... The quantitative calculation of the Hazen coefficient for medium-graded soils was realized, in which... All are experimental calibration parameters, molecular The linear inhibition effect of fine-particle contact blockage in medium-graded soils was quantified; the higher the fine-particle content (FC), the denser the particle contact. The larger the particle size, the stronger its blocking effect on the osmotic channel; the denominator... The nonlinear clogging effect of fine particles filling pores was quantified. Fine particles only adhere to the surface of coarse particles, resulting in a weak clogging effect. However, when the FC (filled pore volume) exceeds a critical value of 10%, fine particles fill the pore channels between coarse particles, leading to a sharp decrease in the cross-sectional area of ​​the permeable channels. Therefore, through… The nonlinear enhancement of the quantification blockage effect is consistent with the physical law that when FC exceeds 30%, the permeability of medium-graded soil is extremely weak, and the Hazen coefficient approaches 0. (Basic permeability term) The effect of fine-grained content in medium-graded soil on the degree of blockage of seepage channels in coarse-grained soil was quantified, and the foundation seepage item was analyzed. The influence of the contact state of coarse soil particles in medium-graded soil on permeability was quantified, where 1 represents the baseline term and the contact effect parameter. A set coefficient representing particle morphology (sphericity / roundness) is used by... Quantitative contact looseness, fine particle mobility coefficient The closer to 0, the looser the particle contact, the larger the pores between particles, and the better the connectivity. The ratio of large particle skeleton to small particle filler is quantified, the denominator quantifies the conduction effect of large particle skeleton, and the molecular weight quantifies the blockage effect of small particle filler.

[0035] Therefore, compared to the conventional Hazen model It comprehensively covers fine-particle clogging, gradation characteristics, and contact effects.

[0036] Furthermore, the process of calculating the fine-grained activity coefficient includes: A first ratio is calculated between the current fine-grained content data and the maximum value of historical fine-grained content data. Based on the average of the maximum and minimum values ​​of historical void ratio data, a coarse-grained soil reference void ratio is generated. A second ratio is calculated between the coarse-grained soil reference void ratio and the current void ratio data. Based on the product of the first and second ratios, the fine-grained activity coefficient is determined. The measured fine-grained content data includes the current fine-grained content data and the maximum value of historical fine-grained content data. The measured void ratio data includes the current void ratio data and historical void ratio data. The process of calculating the dynamic fine-particle blockage coefficient includes: The ratio of the controlled particle size to the effective particle size and the current porosity data are substituted into the attenuation coefficient calibration formula to generate a dynamic fine particle blockage coefficient, wherein the current porosity data includes the controlled particle size and the effective particle size.

[0037] Specifically, the process of calculating the fine particle activity coefficient and the dynamic fine particle blockage coefficient can be expressed as:

[0038] In the formula, Indicates the reference void ratio of coarse-grained soil. These represent the maximum and minimum values ​​of the historical porosity data, respectively. These represent the current fine particle content data and the historical maximum fine particle content data, respectively. This indicates the current porosity data. Indicates the fine-particle activity coefficient. These represent the currently detected control particle size (i.e., the particle size smaller than 60% of the mass fraction of particles in historical measured particle size data) and the effective particle size (i.e., the particle size smaller than 10% of the mass fraction of particles in historical measured particle size data), respectively. This represents the dynamic fine-grained clogging coefficient.

[0039] like Figure 3 As shown, the process for determining the calibration permeability coefficient of moderate soil further includes: The initial intermediate soil calibration permeability coefficient is obtained from Darcy's law of the permeability test. The third ratio of the current temperature hydrodynamic viscosity to the standard temperature hydrodynamic viscosity is calculated. Based on the product of the initial intermediate soil calibration permeability coefficient and the third ratio, the intermediate soil calibration permeability coefficient is generated. The process of generating the dynamic fine-grain content coefficient using the least squares fitting formula includes: The dependent variable is constructed based on the effective particle size, the measured permeability coefficient of medium soil, the dynamic fine-particle blockage coefficient, the measured void ratio data, and the fine-particle activity coefficient. The independent variable is constructed based on the square of the measured fine-particle content data. The fitting formula is constructed based on the fact that the dependent variable is equal to the product of the independent variable and the dynamic fine-particle content coefficient. The fitting formula is solved by the weighted least squares method of minimizing the weighted sum of squared residuals to generate the dynamic fine-particle content coefficient.

[0040] It should be noted that in this embodiment, during the real-time prediction and calculation of the permeability coefficient of medium-coarse soil with graded gradation based on the data collected by sensors in the construction area, it is not necessary to update the seepage velocity obtained from the settlement test and the initial calibrated permeability coefficient of medium soil generated by Darcy's law from the permeability test in real time. It is sufficient to calibrate them as empirical values ​​only in the initial stage.

[0041] Specifically, the process of calculating and determining the standard permeability coefficient and dynamic fine-grain content coefficient of medium-grade soil can be expressed as: In the formula, This represents the initial calibration permeability coefficient of moderate soil at the current temperature. These represent the steady-state seepage flow rate, sample height, sample cross-sectional area, head difference, and measurement time obtained from the Darcy's law in the permeability test, respectively. This indicates the determination of the permeability coefficient of medium-grade soil at a standard temperature of 20℃. These represent the hydrodynamic viscosity at the current temperature and the hydrodynamic viscosity at the standard temperature (20℃), respectively, obtained from a table of physical constants. Indicated as an independent variable The square of the measured fine particle content data, The dependent variable representing the j-th sample ,in It represents the square of the effective particle size (i.e., the particle size smaller than 10% of the mass fraction of particles in the measured particle size data of the sample set). Indicates the dynamic fine-particle clogging coefficient. This represents the measured fine particle content data. Indicates the fine-particle activity coefficient. Indicates the dynamic fine particle content coefficient. This represents the objective function obtained using the weighted least squares method, which is constructed based on the fitting formula. This means that the weighted least squares method will make the sample set... The dynamic fine-grained content coefficient that minimizes the weighted sum of squared residuals is used as the output dynamic fine-grained content coefficient, where m represents the total number of samples and j represents the sample number. Let represent the weight of the j-th sample. The weighted least squares method obtains the dynamic fine-grain content coefficient by taking the derivative of the objective function and setting the derivative to 0.

[0042] Furthermore, the process of generating the permeability coefficient of the second-grade medium-coarse-grained soil includes: The improved USBR model calculates the gradation particle size parameters based on the controlled particle size and effective particle size, calculates the gradation void ratio based on the measured void ratio data and the measured fine particle content data, and generates the permeability coefficient of the second gradation medium-coarse-grained soil based on the gradation particle size parameters and the gradation void ratio. The gradation parameters include gradation particle size parameters and gradation void ratio.

[0043] Specifically, the improved USBR model can be represented as: In the formula, These represent the controlled particle size and the effective particle size, respectively. Indicates the particle size distribution parameter. Indicates the porosity of the graded distribution. This represents the measured void ratio data, where k represents a set coefficient. This represents the measured fine particle content data. This represents the permeability coefficient of medium-coarse-grained soil in the second grade.

[0044] In particular, compared to the conventional USBR model The improved USBR model considers the influence of the gradation particle size parameters and gradation void ratio of medium-graded soil on the permeability coefficient of medium-coarse-grained soil.

[0045] Furthermore, the process of calculating the measured fine particle content data includes: Porosity is calculated based on the conductivity data, the dielectric constant of solid particles, and the dielectric constant of pore water, and the measured fine particle content data is calculated based on the porosity.

[0046] Specifically, the process of calculating the measured fine particle content data can be expressed as follows: In the formula, Represents conductivity data. They represent the dielectric constant of pore water and the dielectric constant of solid particles, respectively. Indicates porosity. This indicates the measured fine particle content data.

[0047] Specifically, the TDR (Time Domain Reflectometer) sensor estimates the soil moisture content by measuring the soil's dielectric constant, and then calculates the measured void ratio based on the moisture content. The process is as follows: In the formula, Indicates moisture content, This represents the dielectric constant data. These are all typical parameters, with a preferred value of -5.3 × 10⁻⁶. −2 2.92×10 −2 -5.5×10 −4 4.3×10 −6 , This represents the measured void ratio data. This represents the measured fine particle content data. This represents the clay correction factor, with a preferred value of -0.02.

[0048] Specifically, the discrete volume distribution of the soil and the corresponding particle size range are determined based on the scattered light intensity distribution data collected by the laser diffraction sensor, and then the measured particle size is calculated.

[0049] like Figure 4 As shown, the process of generating the predicted permeability coefficient of medium-coarse-grained soil with appropriate gradation further includes: Based on the measured void ratio data, measured fine particle content data, measured particle size, permeability coefficient of medium-coarse soil in the first grade and permeability coefficient of medium-coarse soil in the second grade, a measured input list is constructed. The measured input list is passed through multi-scale convolutional units to generate comprehensive features; The comprehensive features are passed through the output mapping layer to generate weighting coefficients, and the permeability coefficients of the first grade medium-coarse-grained soil and the second grade medium-coarse-grained soil are weighted and calculated using the weighting coefficients to generate the predicted permeability coefficient of the grade medium-coarse-grained soil. The weighted model mentioned above includes multi-scale convolutional units and output mapping layers.

[0050] like Figure 4 As shown, the process of generating comprehensive features further includes: The measured input list is passed through a fine-grained scale convolutional layer to generate fine-grained effect features; The measured input list is passed through a gradation scale convolutional layer to generate gradation effect features; The measured input list is passed through a macro-scale convolutional layer to generate comprehensive effect features; The fine-grained effect features, gradation effect features, and combined effect features are vector-concatenated to generate the combined feature; The multi-scale convolutional unit includes fine-grained convolutional layers, graded-scale convolutional layers, and macro-scale convolutional layers.

[0051] Specifically, the process of generating the permeability coefficient of the predicted graded medium-coarse-grained soil can be expressed as follows: In the formula, This represents the actual input list. These represent the effective particle size, control particle size, and median particle size, respectively. The ratio representing the effective particle size. These represent the measured void ratio data and the measured fine particle content data, respectively. These represent the permeability coefficients of the first-grade medium-coarse-grained soil and the second-grade medium-coarse-grained soil, respectively. These represent the characteristics of fine-grained effect, gradation effect, and combined effect, respectively. express Activation function All represent learnable weight coefficients. Both represent learnable bias vectors. Indicates comprehensive characteristics, This indicates vector concatenation, and w represents the weighting coefficient.

[0052] Preferably, the fine-grained effect features, gradation effect features, and comprehensive effect features are flattened into one-dimensional vectors and then concatenated to generate comprehensive features.

[0053] Furthermore, the process of generating the predicted permeability coefficient of medium-coarse-grained soil with appropriate gradation also includes: A weight regularization term is constructed based on the weighting coefficients, and a total loss function is constructed based on the weight regularization term and the mean squared error term. The weighted model is then trained based on the total loss function.

[0054] Specifically, the total loss function can be expressed as: In the formula, These represent the total loss function, the mean squared error term, and the regularization term, respectively. Let represent the first weighting coefficient and the second weighting coefficient of the i-th sample, respectively.

[0055] Specifically, in the experimental process of the intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion described in this embodiment, ice-water deposits of a reservoir dam foundation were selected as medium-graded soil. Twenty groups of soil samples with different gradations were prepared to cover the common gradation variation range in actual engineering. Sieve tests were performed on each group of soil samples to obtain a complete set of gradation characteristic parameters, and basic physical indicators such as dry density and void ratio were measured. The average relative error of the empirical model using only the improved Hazen model was 61.29%, the average relative error of the empirical model using only the improved USBR model was 58.33%, and the average relative error of the weighted model using the improved Hazen model, the improved USBR model, and the combined model was 2.17%. It can be seen that this embodiment effectively solves the problems of low experimental efficiency, poor dynamic adaptability, and insufficient prediction accuracy of the empirical model for medium-graded soil.

[0056] In this embodiment, multi-source sensor data fusion, improved model optimization, and intelligent weighted fusion are used to avoid the problems of traditional coarse-grained soil permeability coefficient prediction methods relying on a single data source, poor model adaptability, and insufficient prediction accuracy. By introducing dynamic fine-grained clogging coefficient and dynamic fine-grained contact coefficient to optimize the Hazen model, and by introducing gradation parameters to optimize the USBR model, both types of models can accurately match the permeability characteristics of medium-graded coarse-grained soil. The core features are extracted by convolutional neural network and adaptive weighted fusion of the results of the two models is performed. This combines the empirical model with intelligent algorithms, effectively solving the problems of low experimental efficiency, poor dynamic adaptability, and insufficient prediction accuracy of the empirical model for medium-graded soil. By improving the Hazen model, the effects of fine-grained blockage, gradation characteristics, and contact effects are comprehensively covered. The Hazen correction coefficient for medium-graded soil is derived by using the ratio of the negative exponential calculation of the dynamic fine-grained blockage coefficient, fine-grained content, and fine-grained activity coefficient to the square term of fine-grained content. This accurately quantifies the correction effect of gradation and the dynamic interaction of fine particles on the permeability coefficient. The fine-grained content blockage term, combined with the blockage degree parameter determined by the settlement test in the construction area, achieves nonlinear quantification of the blockage effect, which conforms to the complex evolution law of fine-grained blockage in actual engineering. The basic permeability term, through the product calculation of contact effect parameter, fine-grained activity coefficient, and measured particle size, fully captures the fundamental influence of particle contact state on permeability, enabling the improved Hazen model to comprehensively adapt to the permeability characteristics of medium-graded coarse-grained soil. The improved USBR model focuses on the core coupling relationship between gradation and porosity. By controlling the particle size and the gradation particle size parameter calculated by the effective particle size, as well as the gradation porosity combined with the measured void ratio and fine-grained content, it achieves accurate adaptation to the permeability characteristics under different gradation conditions. By improving the dual-path prediction of the Hazen model and the improved USBR model, and combining it with a subsequent refined intelligent weighted fusion strategy, the advantages of the prediction results are complemented. Through three convolutional layers of multi-scale convolutional units at the fine-grained scale, gradation scale, and macro scale, the core features of fine-grained dynamic effects, gradation characteristic effects, and comprehensive geological condition effects are accurately extracted, avoiding the one-sidedness of single-scale feature extraction. This achieves adaptive weighted fusion of the results of the two models, meeting the requirements of experimental efficiency and prediction accuracy for medium-graded soils during construction.

[0057] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0058] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0059] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart prediction method for the permeability coefficient of coarse-grained soil based on multi-data fusion, characterized in that, include: The measured porosity data is calculated based on the dielectric constant data collected by the TDR sensor in the construction area, the measured fine particle content data is calculated based on the conductivity data collected by the conductivity sensor in the construction area, and the measured particle size is calculated based on the scattered light intensity distribution data collected by the laser diffraction sensor in the construction area. The measured void ratio data, measured fine particle content data, and measured particle size are used to generate the permeability coefficient of medium-coarse-grained soil in the first grade by using an improved Hazen model, wherein the improved Hazen model includes dynamic fine particle blockage coefficient and dynamic fine particle contact coefficient. The measured void ratio data, measured fine particle content data, and measured particle size are used to generate the permeability coefficient of the second-grade medium-coarse-grained soil by using an improved USBR model, wherein the improved USBR model includes gradation parameters. The permeability coefficients of the first and second graded medium-coarse-grained soils are used to generate the predicted permeability coefficients of the graded medium-coarse-grained soils through a weighted model based on a convolutional neural network architecture.

2. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 1, characterized in that, The process of generating the permeability coefficient of medium-coarse-grained soil in the first grade includes: The fine particle activity coefficient is calculated based on the measured porosity data and the measured fine particle content data, and the dynamic fine particle blockage coefficient is calculated based on the measured fine particle content data and the measured porosity data. The calibration permeability coefficient of medium-strength soil was determined by conducting permeability tests based on Darcy's law with temperature correction in the construction area. Based on the measured fine particle content data, measured particle size, fine particle activity coefficient, measured permeability coefficient of medium soil and dynamic fine particle blockage coefficient, the dynamic fine particle content coefficient is generated by least squares fitting method. The measured fine particle content data, measured particle size, fine particle activity coefficient, dynamic fine particle blockage coefficient, and dynamic fine particle content coefficient are substituted into the improved Hazen model to generate the permeability coefficient of the medium-coarse-grained soil in the first grade.

3. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 2, characterized in that, The process of substituting the improved Hazen model into the permeability coefficient of the first-grade medium-coarse-grained soil includes: The improved Hazen model calculates the negative exponent value of the product of the dynamic fine-particle blockage coefficient, the measured fine-particle content data, and the fine-particle activity coefficient, calculates the product of the dynamic fine-particle content coefficient and the square of the measured fine-particle content data, and generates the Hazen correction coefficient for graded medium soil based on the ratio of the negative exponent value to the product. The improved Hazen model calculates an index value based on measured fine particle content data and a blockage nonlinearity index. The fine particle content blockage term is generated by multiplying the index value and a blockage degree parameter, wherein the blockage degree parameter is determined by settlement tests in the construction area. The improved Hazen model is based on the product of contact effect parameters, fine particle activity coefficient, and measured particle size to generate the basic permeability term; The improved Hazen model is based on the product of the Hazen correction coefficient for medium-graded soil, the fine-grained content blocking term, and the basic permeability term to generate the permeability coefficient for medium-coarse-grained soil of the first grade.

4. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 2, characterized in that, The process of calculating the fine-grained activity coefficient includes: A first ratio is calculated between the current fine-grained content data and the maximum value of historical fine-grained content data. Based on the average of the maximum and minimum values ​​of historical void ratio data, a coarse-grained soil reference void ratio is generated. A second ratio is calculated between the coarse-grained soil reference void ratio and the current void ratio data. Based on the product of the first and second ratios, the fine-grained activity coefficient is determined. The measured fine-grained content data includes the current fine-grained content data and the maximum value of historical fine-grained content data. The measured void ratio data includes the current void ratio data and historical void ratio data. The process of calculating the dynamic fine-particle blockage coefficient includes: The ratio of the controlled particle size to the effective particle size and the current porosity data are substituted into the attenuation coefficient calibration formula to generate a dynamic fine particle blockage coefficient, wherein the current porosity data includes the controlled particle size and the effective particle size.

5. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 3, characterized in that, The process of determining the calibration permeability coefficient of medium-grade soil includes: The initial intermediate soil calibration permeability coefficient is obtained from Darcy's law of the permeability test. The third ratio of the current temperature hydrodynamic viscosity to the standard temperature hydrodynamic viscosity is calculated. Based on the product of the initial intermediate soil calibration permeability coefficient and the third ratio, the intermediate soil calibration permeability coefficient is generated. The process of generating the dynamic fine-grain content coefficient using the least squares fitting formula includes: The dependent variable is constructed based on the effective particle size, the measured permeability coefficient of medium soil, the dynamic fine-particle blockage coefficient, the measured void ratio data, and the fine-particle activity coefficient. The independent variable is constructed based on the square of the measured fine-particle content data. The fitting formula is constructed based on the fact that the dependent variable is equal to the product of the independent variable and the dynamic fine-particle content coefficient. The fitting formula is solved by the weighted least squares method of minimizing the weighted sum of squared residuals to generate the dynamic fine-particle content coefficient.

6. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 4, characterized in that, The process of generating the permeability coefficient of the second-grade medium-coarse-grained soil includes: The improved USBR model calculates the gradation particle size parameters based on the controlled particle size and effective particle size, calculates the gradation void ratio based on the measured void ratio data and the measured fine particle content data, and generates the permeability coefficient of the second gradation medium-coarse-grained soil based on the gradation particle size parameters and the gradation void ratio. The gradation parameters include gradation particle size parameters and gradation void ratio.

7. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 1, characterized in that, The process of generating the predicted permeability coefficient of medium-coarse-grained soil includes: Based on the measured void ratio data, measured fine particle content data, measured particle size, permeability coefficient of medium-coarse soil in the first grade and permeability coefficient of medium-coarse soil in the second grade, a measured input list is constructed. The measured input list is passed through multi-scale convolutional units to generate comprehensive features; The comprehensive features are passed through the output mapping layer to generate weighting coefficients, and the permeability coefficients of the first grade medium-coarse-grained soil and the second grade medium-coarse-grained soil are weighted and calculated using the weighting coefficients to generate the predicted permeability coefficient of the grade medium-coarse-grained soil. The weighted model mentioned above includes multi-scale convolutional units and output mapping layers.

8. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 7, characterized in that, The process of generating comprehensive features includes: The measured input list is passed through a fine-grained scale convolutional layer to generate fine-grained effect features; The measured input list is passed through a gradation scale convolutional layer to generate gradation effect features; The measured input list is passed through a macro-scale convolutional layer to generate comprehensive effect features; The fine-grained effect features, gradation effect features, and combined effect features are vector-concatenated to generate the combined feature; The multi-scale convolutional unit includes fine-grained convolutional layers, graded-scale convolutional layers, and macro-scale convolutional layers.

9. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to claim 7, characterized in that, The process of generating the predicted permeability coefficient of medium-coarse-grained soil with appropriate gradation also includes: A weight regularization term is constructed based on the weighting coefficients, and a total loss function is constructed based on the weight regularization term and the mean squared error term. The weighted model is then trained based on the total loss function.

10. The intelligent prediction method for permeability coefficient of coarse-grained soil based on multi-data fusion according to any one of claims 1 to 9, characterized in that, The process of calculating the measured fine particle content data includes: Porosity is calculated based on the conductivity data, the dielectric constant of solid particles, and the dielectric constant of pore water, and the measured fine particle content data is calculated based on the porosity.

Images