Method for decoupling deformation of expansive soil during freezing process based on machine learning factor analysis

Abstract

The application discloses an expansive soil freezing process deformation decoupling method based on machine learning factor analysis, which comprises the following steps: placing a prepared soil sample into a frost heaving testing machine, and measuring the parameter changes of the soil sample in the freezing process when the temperature decreases; and importing the obtained data into MATLAB, and using the ICA and regression fitting algorithm programs therein to obtain the volume increase or decrease caused by the volume ice content rate change and the volume water content rate change caused by the unfrozen water, respectively. The method can obtain the volume changes caused by water freezing and soil particle water loss, respectively, on the premise that the expansive soil freezing process includes the volume increase caused by water freezing into ice and the volume decrease caused by the shrinkage of the expansive soil particles due to water loss. The application can distinguish the volume changes of the soil body under the influence of different factors, and can also accurately obtain the function relationship formula and the relationship curve between the volume change and the generating factors by using machine learning factor analysis and regression fitting.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of volume change decoupling of expansive soil frost heaving process, and relates to a machine learning decoupling method, in particular to an expansive soil frost heaving process deformation decoupling method based on machine learning factor analysis. BACKGROUND

[0002] In signal processing, independent component analysis (ICA) is a computational method for separating a multivariate signal into additive subcomponents. This is done by assuming that the subcomponents are non-Gaussian signals and are statistically independent of each other. ICA is a special case of blind source separation. A common example application of independent component analysis (also known as factor analysis) is the "cocktail party problem" of listening to one person's speech in a noisy room.

[0003] Machine learning regression fitting is to use certain basis functions to represent known data by linear combination thereof, for fitting the mutual functional relationship of the obtained data. Among them, the basis function can select sigmoid function. Sigmoid function is also called Logistic function, the value range is (0, 1), which can be used for binary classification, and the effect is better when the feature difference is relatively complex or not particularly large.

[0004] The existing research on the volume change law of soil frost heaving is still limited to obtaining various relationship curves and change curves through experimental measurement data, and the functional relationship between soil volume change and influencing factors cannot be accurately and effectively obtained. SUMMARY

[0005] The purpose of the application is to provide an expansive soil frost heaving process deformation decoupling method based on machine learning factor analysis. The method understands that in the process of expansive soil freezing, there is both volume increase caused by water freezing into ice and volume decrease caused by shrinkage of expansive soil particles due to water loss.

[0006] The purpose of the application is achieved by the following technical solutions:

[0007] An expansive soil frost heaving process deformation decoupling method based on machine learning factor analysis comprises the following steps:

[0008] Step 1: Prepare N groups of expansive soil samples with the same size but different volume water content, N>=3, and wrap the expansive soil samples with preservative bags before placing them in a frost heaving testing machine to prevent water loss;

[0009] Step two, insert temperature sensor, time domain reflectometry conductive probe and neutron scattering instrument detector in the expansive soil sample for determining the volume water content of all water in the soil, use temperature sensor, time domain reflectometry conductive probe and neutron scattering instrument detector to measure the temperature change, the volume water content change of unfrozen water and the volume water content change of all water of the expansive soil sample in the frost heaving testing machine respectively; install laser range finder in the frost heaving testing machine, use laser range finder to measure the height and diameter change of the expansive soil sample in the testing machine;

[0010] Step three, set the minimum frost heaving temperature as-18℃, wait for the temperature in the testing machine to decrease to 3℃, start to record data every 5 minutes until the temperature of the soil sample decreases to-18℃, end the recording process, wherein:

[0011] The data includes temperature T j,i , volume water content change of unfrozen water Δω j,i , volume water content change of all water Δa j,i , volume ice content change Δθ j,i , height change of the soil sample ΔH j,i and diameter change of the soil sample ΔD j,i ;

[0012] The volume ice content change Δθ j,i is:

[0013] Δθ j,i = Δa j,i - Δω j,i

[0014] The vector composed of n groups of height change values of the jth soil sample is:

[0015]

[0016] The vector composed of n groups of diameter change values of the jth soil sample is:

[0017]

[0018] The vector composed of n groups of volume ice content change values of the jth soil sample is:

[0019]

[0020] The vector composed of n groups of volume water content change values of unfrozen water of the jth soil sample is:

[0021]

[0022] ​​​​where j represents the jth soil sample, j = 1, 2,..., N, and i represents the ith record, i = 1, 2,..., n.

[0023] Step four, the recorded data is inputted into MATLAB, and the volume change caused by the change of volume ice content and the change of volume water content of unfrozen water respectively is calculated by using ICA algorithm, wherein:

[0024] The volume change caused by the change of volume ice content ΔV θ is:

[0025]

[0026] The volume change caused by the change of volume water content ΔV ω is:

[0027]

[0028] The vector composed of the volume changes caused by the change of volume ice content of the n groups of the jth soil sample respectively is:

[0029]

[0030] The vector composed of the volume changes caused by the change of volume water content of the n groups of the jth soil sample respectively is:

[0031]

[0032] Step five, the and are respectively taken as dependent variables, the and are respectively taken as independent variables, and the function relationship between the change of volume ice content and the volume change caused by the change of volume ice content and the function curve graph corresponding thereto are obtained by using regression fitting, wherein:

[0033] The expression of the fitting function relationship between the change of volume ice content of the jth soil sample and the volume change caused by the change of volume ice content is:

[0034]

[0035]

[0036] The sigmoid function of the change of volume ice content Δθ j,i is: [1+exp(Δθ j,i )] -1

[0037] The fitting function relationship expression between the volume change of the jth soil sample and the volume change caused by the volume change of the unfrozen water content is as follows:

[0038]

[0039]

[0040] sigmoid (Δω j,i ) = [1 + exp (Δω j,i )] -1

[0041] In the formula, W is an n*n matrix, and W = w i,j , w m,n = 0, n > m, i, m refer to the ith, mth row, and j, n refer to the jth, nth column.

[0042] Compared with the prior art, the present application has the following advantages:

[0043] 1. The present application uses machine learning factor analysis and regression fitting to distinguish the volume change of the soil under the influence of different factors, and to obtain the function relationship expression and the relationship curve between the volume change and the generating factors.

[0044] 2. The present application combines the related algorithms and applications of indoor soil test and machine learning, and can overcome the shortcomings of the traditional method, i.e., the relationship expression between the volume change of the soil and the influencing factors is not clear and the subsequent calculation and analysis is complex, so that the related function expression and clear curve relationship can be accurately and efficiently obtained. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The flowchart of the swelling soil frost heaving deformation decoupling method of the machine learning factor analysis of the present application. DETAILED DESCRIPTION

[0046] The technical solutions of the present application will be further described below in conjunction with the drawings, but are not limited thereto, and any modification or equivalent replacement to the technical solutions of the present application without departing from the spirit and scope of the present application shall be covered in the protection scope of the present application.

[0047] The present application provides a swelling soil freezing process deformation decoupling method of machine learning factor analysis, which puts the prepared soil sample into a frost heaving test machine to measure the changes of various parameters during the freezing process of the soil sample with temperature reduction; and imports the obtained data into MATLAB to obtain the volume increase or decrease caused by the volume change of the volume ice content and the volume change of the unfrozen water content, respectively, by using the ICA and regression fitting algorithm program therein. Figure 1 As shown in the formula, the method comprises the following steps:

[0048] Step 1: Prepare three expansive soil samples of the same size but different volumetric water content, and label them as soil sample 1, soil sample 2 and soil sample 3 respectively. The soil sample height is 120mm, the diameter is 50mm, and the volumetric water content ω is 0.10, 0.12 and 0.14 respectively.

[0049] Step 2: Wrap the three prepared soil samples in plastic bags and place them in a frost heave testing machine. Insert a temperature sensor, a time domain reflectometer probe, and a neutron scattering detector into the soil samples to measure the soil temperature, the change in volumetric water content of unfrozen water, and the change in volumetric water content of all water. Install a laser rangefinder in the testing machine to measure the change in diameter and height of the soil samples.

[0050] Step 3: Set the minimum frost heave temperature of the frost heave tester to -18℃, adjust the equipment, turn on the tester to cool down, and observe the temperature change. Wait for the lowest soil sample temperature to drop to 3℃.

[0051] Step 4: Record parameters every 5 minutes, including temperature T. j,i Change in volumetric water content of unfrozen water Δω j,i The change in the volumetric water content of all water Δa j,i Change in volumetric ice content Δθ j,i Soil sample height change ΔH j,i and the change in soil sample diameter ΔD j,i Recording was stopped when the temperature of all three soil samples dropped to -18℃.

[0052] In this step, j represents the j-th soil sample, j = 1, 2, 3, and i represents the i-th record, i = 1, 2, ..., ..., n; the change in volumetric ice content is the change in volumetric water content of all water minus the change in volumetric water content of unfrozen water; ΔH j,i and ΔD j,i The units are all cm, T j,i The unit is ℃; Equation (1) represents the vector composed of the height change values ​​of the nth group of the j-th soil sample. Equation (2) represents the vector composed of the diameter variation values ​​of the nth group of the j-th soil sample. Equation (3) represents the vector composed of the volumetric ice content change values ​​of the j-th soil sample in the nth group. Equation (4) represents the vector composed of the changes in the volumetric water content of the unfrozen water in the nth group of the j-th soil sample.

[0053]

[0054]

[0055]

[0056]

[0057] Step 5, statistics of recorded data, import it into MATLAB in the form of matrix m file, use ICA algorithm program to calculate the volume change caused by the volume ice content rate change and the volume change caused by the unfrozen water volume water content rate change respectively.

[0058] In this step, the volume change ΔV θ The volume change ΔV ω The vector composed of the volume change caused by the volume ice content rate change of the jth soil sample n groups respectively is expressed by formula (6) The vector composed of the volume change caused by the unfrozen water volume water content rate change of the jth soil sample n groups respectively is expressed by formula (7) The vector composed of the volume change caused by the unfrozen water volume water content rate change of the jth soil sample n groups respectively is expressed by formula (7) And The unit of 3 .

[0059]

[0060]

[0061]

[0062]

[0063] Step 6, take And As dependent variables respectively, take And As independent variables respectively, use the regression fitting algorithm to fit the function relationship and get the corresponding curve figure, wherein: the base function selected by the regression fitting is sigmoid function.

[0064] In this step, the fitting function relationship expression between the volume ice content rate change of the jth soil sample and the volume change caused by it is expressed by formula (9), the base function selected in formula (9) is expressed by formula (10), and the expression of formula (10) is expressed by formula (11); the fitting function relationship expression between the unfrozen water volume water content rate change of the jth soil sample and the volume change caused by it is expressed by formula (12), the base function selected in formula (12) is expressed by formula (13), and the expression of formula (13) is expressed by formula (14).

[0065]

[0066]

[0067] sigmoid(Δθj,i ) = [1 + exp(Δθ j,i )] -1 (11)

[0068]

[0069]

[0070] sigmoid(Δω j,i ) = [1 + exp(Δω j,i )] -1 (14)

[0071] where W is an n x n matrix, which can be denoted as W = w i,j , w m,n = 0, n > m, i, m refer to the i, mth row; j, n refer to the j, nth column.

Claims

1. A method for decoupling deformation of expansive soil during freezing process based on machine learning factor analysis, characterized in that The method comprises the following steps: Step one, making a swelling soil sample of the same size but different volumetric water content, The swelling soil sample is wrapped with a preservative bag and placed in the frost heaving testing machine. Step two, inserting temperature sensor, time domain reflectometry conductive probe and neutron scattering instrument detector in the expansive soil sample for determining the volume water content of all water in the soil, using temperature sensor, time domain reflectometry conductive probe and neutron scattering instrument detector to respectively measure the temperature change, the change of the volume water content of unfrozen water and the change of the volume water content of all water of the expansive soil sample in the frost heaving testing machine; installing laser range finder in the frost heaving testing machine, and using the laser range finder to measure the change of the height and diameter of the expansive soil sample in the testing machine; Step three, setting the minimum frost heaving temperature as-18℃, waiting for the temperature in the testing machine to decrease to 3℃, and then recording data every 5 minutes until the temperature of the soil sample decreases to-18℃, and then ending the recording process; Step four, inputting the recorded data into MATLAB, and using ICA algorithm to calculate the volume change caused by the change of the volume ice content and the change of the volume water content of unfrozen water respectively. Step 5: and As dependent variables respectively, and Using these variables as independent variables, regression fitting was used to obtain the functional relationships and corresponding function curves between the changes in volumetric ice content and unfrozen water volumetric water content and the resulting volumetric changes, respectively. The first soil sample The vector consisting of the volume change caused by the change in ice content of each recorded volume is: ； The first Soil sample The vector consisting of the volume change caused by each of the changes in the volume content of the unfrozen water is: The first ； The first soil sample vector of volume ice fraction change values is: The first soil sample vector of volume water content change values for the non-frozen water is: In the formula, indicates the th soil sample, , indicates the th record, .

2. The machine learning factor analysis based method for decoupling of swelling soil freeze process deformation of claim 1, wherein The data in step three includes temperature volume water content change of unfrozen water volume water content change of all water volume ice content change soil sample height change soil sample diameter change wherein: Volume ice fraction change Is: The first soil sample vector of height change values is: The first soil sample vector of diameter change values is: 。 3. The machine learning factor analysis based method for decoupling of swelling soil freeze process deformation of claim 1, wherein In step four, the volume change caused by the change in ice fraction Is: ； Volume change due to change in volumetric moisture content Is: 。 4. The machine learning factor analysis based method for decoupling of swelling soil freeze process deformation of claim 1, wherein In the fifth step, the first The fitting function expression between the ice content rate change of the soil sample volume and the volume change caused by it is as follows: No. The fitting function expression for the relationship between the change in unfrozen water volumetric moisture content of a soil sample and the resulting volumetric change is: 。

Images