Method for predicting soil-water characteristic relation of soil body based on big data analysis

Abstract

The invention provides a big data analysis-based soil mass soil-water characteristic relation prediction method, which comprises the following steps of: firstly, establishing a description model of soil mass pore size distribution based on a fractal geometry method, and further establishing a soil mass capillary suction model considering pore fractal characteristics by combining a capillary theory; the theoretical model realizes rapid and reliable prediction of the soil-water characteristic relation under the condition of a known soil body pore fractal structure. According to the model prediction method, by means of a big data analysis means, soil-water characteristic curves actually measured by a large number of soil samples are collected, soil pore structure fractal characteristics are reversely deduced by means of a theoretical model, and the corresponding relation between soil body apparent physical indexes and pore fractal parameters is established. And the pore fractal structure can be quickly predicted based on any soil sample apparent physical index. In combination with the theoretical model and the prediction method, the defects of long test time, high cost and the like of the traditional method are finally overcome, and rapid and reliable prediction of the soil-water characteristic curve is realized.

Description

Technical Field

[0001] The present invention relates to the fields of soil science, data analysis and prediction technology, and in particular to a soil-water characteristic curve prediction method based on big data analysis. Background Art

[0002] The soil-water characteristic relationship is a characteristic relationship curve that describes the relationship between soil water saturation and soil water suction. This relationship is the basis for the study of slope seepage and control in water conservancy projects, soil hydrodynamic properties, soil water regulation and utilization, soil improvement in agricultural engineering, etc. Therefore, the measurement and prediction of soil-water characteristic curves have an important position and practical significance in the field of geotechnical engineering.

[0003] Traditional methods for determining soil-water characteristic curves include direct measurement methods and indirect prediction methods. Direct test methods usually rely on laboratory tests, such as pressure plate method, centrifugation method, axis translation technology, etc. These direct test methods take a lot of time and are also very costly. They are also greatly affected by test equipment and operators and have many uncertainties. The indirect prediction method predicts soil-water characteristic curves through mathematical models, including empirical formula method, soil conversion function method, physical empirical model, etc. The indirect prediction method overcomes the time-consuming, labor-intensive and uneconomical problems of measuring the curve through experiments, and becomes a simpler, faster and more effective method. However, most of these mathematical model parameters lack clear physical meanings.

[0004] The soil-water characteristic curve is essentially a macroscopic reflection of the soil pore structure, so some scholars have attempted to establish a mathematical model based on the soil pore structure to predict the soil-water characteristic curve. However, the pore structure is affected by many factors such as density and particle grading, and accurate prediction remains a challenge until now. Summary of the invention

[0005] In view of the above situation, the present invention provides a soil water characteristic curve prediction method, device and equipment based on big data analysis to achieve fast and accurate prediction. The method is divided into two parts. First, a soil water characteristic curve prediction model based on pore structure distribution is established, and then the corresponding relationship between pore size distribution and dry density, particle grading and other indicators is obtained by big data statistics, and finally the soil water characteristic curve is predicted quickly and accurately based on the apparent physical indicators of the soil (such as dry density and particle grading). The technical solution of the present invention is as follows: A soil-water characteristic curve prediction method based on big data analysis includes the following steps: Step 1: Establish a soil-water characteristic curve prediction model based on pore structure distribution. The model is characterized by using the Menger sponge model, a porous fractal model commonly used in geotechnical media, to establish a three-dimensional fractal description of the soil pore structure. The specific construction method includes: ① First, assume that the soil has an initial side length of L Cube; ② The cube is m After dividing equally, there is m 3 sub-cubes; randomly remove a certain number of cubes, and the remaining j ③ Continue to divide and remove all the remaining cubes in the same way. At this time, the number of remaining small cubes increases and their sizes decrease. ④ When the operation is completed k The number of cubes is j k , the side length is R = L / m k For determining soil samples, parameters j , k , m are all constants, and their fractal dimensions are D = lg( j ) / lg( m ).

[0006] As an improvement, the total volume of the soil in the three-dimensional sponge model V t The solid volume of soil V s Total pore volume V p Composition, total volume of soil V t , soil solid volume V s Total pore volume V p They are: (1); In the formula, R is the minimum pore radius of the soil sample, L is the computational domain size, D is the fractal dimension.

[0007] In the process of soil dehumidification, large pores in the soil always lose water first, while small pores lose water later. If the pore radius of the water-air interface is r , it can be considered that all pores larger than this pore size are filled with air, and all pores smaller than this pore size are filled with water, then the pore size greater than or equal to is r The pore volume (i.e. the volume filled with air) Vp (≥ r ) accounts for the total volume, expressed as (2); Air filled volume V p (≥ r ) is expressed as the content of the total volume: (3); In the formula V p (≤ r ) represents the volume of water in the soil, V s is the volume of soil particles, V t is the total volume of the soil. The volume of the material is closely related to its mass. The physical quantities involved in the above formula are expressed as: (4); Where: m w is the water mass in the soil, m s is the mass of soil particles, d s is the relative density of soil particles; ρ w is the density of water; ρ d is the soil dry density, w is the soil mass moisture content.

[0008] Combining equations (2), (3) and (4), we can obtain: (5); Taking the logarithm of both sides of the above equation, the soil mass moisture content is w With critical radius r The relationship is expressed as (6); As an improvement, an ideal capillary tube is used to describe the pores in the soil pore fractal model, and the capillary suction is established using the Young-Laplace equation. S Size and pore radius r The contact angle of the solid-liquid-gas three-phase interface θ Relationship (7); Where: r is the pore radius, S is the capillary suction force, T s is the surface tension of the liquid.

[0009] Combining equation (6) with equation (7), we can get ρ d Capillary suction under influence S Soil mass moisture content w The relationship between (8); The prediction model includes T s , θ , L , d s , D , ρ d , a total of 6 unknown parameters: T s is the surface tension, which is 0.073 N / m at 25°C; θ is the contact angle, which is equal to 0 during the dewetting process; d s is the relative density of the soil, measured by the pycnometer method; dry density ρ d Measured according to the oven-drying method.

[0010] Step 2: Establish soil pore fractal characteristic parameters (fractal dimension) based on BP neural network D and maximum aperture L )The corresponding relationship between pore size distribution and dry density, particle grading and other indicators. The specific methods include: As an improvement, a large number of soil moisture content data from previous experiments were collected. w ~Soil water suction S Curve data and its corresponding soil physical index parameters, such as relative density, dry density, particle grading, etc., and establish their database; As an improvement, the soil pore size characteristic distribution is plotted based on the data in the database and combined with the soil fractal model. The specific steps include: ① According to formula (8), the left term of the formula As the vertical axis, the right side ② Draw a data point graph with y as the horizontal axis and perform linear fitting on the data points; ② Obtain the slope based on the linear fitting results and the intercept , the fractal dimension can be calculated D and maximum aperture L .

[0011] As an improvement, for different dry densities in the database ρ d , particle grading dThe soil samples were preprocessed, including cleaning, denoising, and normalization, and the data set was divided into a training set and a test set.

[0012] As an improvement, the topological structure of the BP neural network is determined, with the soil surface physical parameters as input and the pore fractal parameters as output, and the number of hidden layer neurons is determined according to the empirical formula: (9); In the formula, n is the number of neurons in the hidden layer, m is the number of nodes in the input layer, l is the number of nodes in the output layer, k ∈(0,10).

[0013] As an improvement, the initial weights and thresholds are input into the neural network for training, and the error between the real data and the predicted data of the pore fractal parameters is calculated. When the error value does not meet the preset accuracy requirement, feedback is given from the output layer to the input layer layer by layer, and the connection weights and thresholds between the layers are continuously adjusted until the error reaches the set stopping condition.

[0014] As an improvement, the test set is run according to the optimized connection weights and thresholds to verify the prediction accuracy of the trained BP neural network.

[0015] The present invention has the following beneficial effects: 1. The model proposed in the present invention is based on fractal theory to establish a description of the soil pore structure, which overcomes the shortcomings of unclear physical meanings and difficult value selection of parameters in traditional models; the characteristic parameters involved in the model have clear physical meanings and can be quickly measured through simple experiments; further combined with the capillary model (Young-Laplace equation) to establish a capillary suction prediction model, it can achieve accurate and reliable prediction of capillary suction.

[0016] 2. Collect a large amount of measured data from soil capillary suction tests, establish a soil capillary suction test database, and conduct statistical analysis of big data in combination with the soil pore structure model to establish model characteristic parameters (fractal dimension D , Maximum aperture L ) and the apparent physical quantities of the soil (dry density, particle grading, etc.), thus realizing the rapid prediction of the soil pore structure; compared with the traditional direct measurement method of the soil pore structure, this prediction method is not only faster but also almost cost-free.

[0017] 3. In the statistical analysis of a large amount of data, the BP neural network intelligent algorithm is further used to conduct in-depth mining of the collected data, overcoming the difficulty of coupling influence analysis of multiple physical factors (dry density, particle grading, relative density, plasticity index, etc.) in traditional data analysis, and realizing the analysis of model characteristic parameters (fractal dimension, etc.) under the influence of multiple physical factors. D , Maximum aperture L ) and also realizes the quantitative evaluation of the importance of each factor.

[0018] 4. By combining the capillary suction prediction model and the pore structure prediction method, it is possible to quickly and reliably predict the soil-water characteristic relationship by relying solely on the apparent physical index parameters of the soil. Compared with the traditional direct testing method, this method can obtain a large amount of soil-water characteristic relationships quickly, so as to provide timely guidance for the project. Compared with the traditional indirect prediction method, this model and method have clear physical meanings, and the prediction results are more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the method for predicting soil-water characteristic curve based on pore structure of the present invention; Figure 2 A flow chart for establishing soil fractal model parameters and apparent physical index database for the present invention; Figure 3 The soil-water characteristic curve data graph of the sample collected by the present invention; Figure 4 It is the soil pore fractal characteristic relationship diagram of the present invention; Figure 5 It is a flow chart of the BP neural network algorithm of the present invention; Figure 6 This is a comparison chart of the true value and predicted value of the fractal dimension of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with the accompanying drawings and examples, but the embodiments of the present invention are not limited thereto. Example 1 In the present invention, in order to facilitate the description of the model and method of the present invention, the soil-water characteristic curve prediction of loess is taken as an example. The specific steps are as follows: Figure 1 As shown, it includes 4 steps from S1 to S4.

[0021] Step S1: Establish a database of the corresponding relationship between soil surface physical indicators (dry density, particle size distribution) and pore structure fractal parameters. This step can be specifically decomposed into S101~S105. Figure 2-4 shown.

[0022] Sub-step S101: Collect the previously measured soil mass moisture content ~ capillary suction curve data, such as Figure 3 shown.

[0023] Sub-step S102: Collect the soil physical index parameters corresponding to step S101, such as dry density, particle gradation, relative density of soil particles, etc., as shown in Table 1.

[0024] Table 1 Physical parameters of soil samples

[0025] Sub-step S103: Substitute the soil mass moisture content ~ capillary suction data into formula (8), and use log( S ) is the horizontal axis, log[ w+1 / d s ] is the ordinate axis, and loess data points are plotted respectively, such as Figure 4 shown.

[0026] Sub-step S104: further performing linear fitting on the data points ( Figure 4 ), get the slope of the fitting function ) and the intercept , calculate the fractal dimension D and maximum aperture L , as shown in Table 2.

[0027] Table 2 Fractal characteristic parameters of loess samples

[0028] Sub-step S105: Collect a large amount of measured data of soil samples, repeat steps S101 to S104, and establish the corresponding relationship between the fractal dimension of each soil sample and the physical index of the soil, as shown in Table 2, dry density ~ fractal dimension / maximum pore size. Based on the large amount of measured data collected, a database is established.

[0029] Step S2: Use BP neural network to train and test the data set, and finally obtain the BP neural network prediction model that best suits the data set, using fractal dimension D For example, this step can be specifically decomposed into S201-S204.

[0030] Sub-step S201: pre-processing the original data in the database, including cleaning, denoising, normalization, etc., and dividing the data set into a training set and a test set in a ratio of 4:1.

[0031] Sub-step S202: Take the soil surface physical parameters as the input layer, including soil particle relative density, dry density, plasticity index, effective particle size, median particle size and limit particle size, a total of 6 characteristic parameters, and use fractal dimension as the input layer. DAs the output layer, the range of hidden layer neurons is determined to be 3 to 12 according to formula (9).

[0032] Sub-step S203: Based on this range, the neural network is run for trial calculation to compare the corresponding mean square error values ​​of different hidden layer neurons. When the number of hidden layer neurons is 11, the mean square error value of the BP neural network is the smallest, thereby determining that the optimal hidden layer neurons is 11.

[0033] Sub-step S204: Determine the coefficient R 2 , mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate the prediction accuracy of the model and select the optimal model.

[0034] Step S3: Obtain the apparent physical indicators of the soil to be tested through simple experiments. For example, the basic physical properties of a loess sample from a certain place are as follows: Table 3 Basic physical properties of soil samples

[0035] Step S4: Taking the soil surface physical index obtained from the test as input, the pore structure fractal parameters of the soil are predicted using the BP neural network prediction model trained in step S2. D , L After prediction, the fractal dimension of the soil sample was obtained as D=2.8152 and the maximum pore size L=0.3524.

[0036] Step S5: The predicted fractal model parameters D , L Substituting into formula (8), the soil-water characteristic curve of the soil sample to be tested can be established. Figure 6 It can be seen that for the 43 soil samples collected, the fractal dimension change trend predicted by the BP neural network is basically consistent with the general trend of the actual fractal dimension curve, and the predicted values ​​of the fractal dimension of most samples are almost consistent with the true values. Therefore, the BP neural network can well predict the fractal dimension of the data set.

[0037] The technical solution of the present invention is explained through the above embodiments, but the present invention is not limited to the above embodiments, that is, it does not mean that the present invention must rely on the above specific embodiments to be implemented. Any improvements made by technicians in the relevant field based on the present invention, or equivalent replacement of the materials selected by the present invention, etc., fall within the scope of protection of the patent.

Claims

1. A soil-water characteristic relationship prediction method based on big data analysis, characterized in that: The following steps are involved: Step 1: Establish a soil-water characteristic curve prediction model based on pore structure distribution. The Menger three-dimensional sponge model, a porous fractal model commonly used in geotechnical media, is used to establish a three-dimensional fractal description of the soil pore structure. Further, combined with the Yang-Laplace equation, a formula for predicting capillary suction based on the soil pore structure is proposed. Step 2, establishing the corresponding relationship between soil pore fractal characteristic parameters and soil surface physical indicators based on BP neural network; Step 3, test the apparent physical indicators of the soil to be tested, use the relationship obtained in step 2 to predict the fractal parameters of the soil pores, and further combine step 1 to predict the soil-water characteristic curve.

2. The soil-water characteristic relationship prediction method based on big data analysis according to claim 1 is characterized in that: The specific construction method of the three-dimensional sponge model in step 1 comprises: 1.1 Assume that the soil has an initial side length of L Cube; 1.2 Place the cube m After dividing into equal parts, there is m 3 sub-cubes; randomly remove a certain number of cubes, and the remaining j small cubes; 1.3 Continue to divide and remove all remaining cubes in the same way. At this time, the number of remaining small cubes increases and their sizes decrease. 1.4 When the operation is carried out k The number of cubes is j k , the side length is R = L / m k , for determining soil samples, parameters j , k , m are all constants, and their fractal dimensions are D = lg( j ) / lg( m ).

3. The soil-water characteristic relationship prediction method based on big data analysis according to claim 2 is characterized in that: Total volume of soil in the 3D sponge model V t The solid volume of soil V s Total pore volume V p composition: （1）； In the formula, R is the minimum pore radius of the soil sample, L is the computational domain size, D is the fractal dimension.

4. The soil-water characteristic relationship prediction method based on big data analysis according to claim 2 is characterized in that: In the three-dimensional sponge model, the pore radius of the water-air interface is r , pores larger than this diameter are all filled with air, while pores smaller than this diameter are all filled with water, and pores larger than or equal to this diameter are all filled with water. r The pore volume is the volume filled with air. V p (≥ r ) is expressed as the content of the total volume: （2）； Air filled volume V p (≥ r ) is expressed as the content of the total volume: （3）； In the formula V p (≤ r ) represents the volume of water in the soil, V s is the volume of soil particles, V t is the total volume of the soil. The volume of the material is closely related to its mass. The physical quantities involved in the above formula are expressed as （4）； In the formula m w is the water mass in the soil, m s is the mass of soil particles, d s is the relative density of soil particles; ρ w is the density of water; ρ d is the soil dry density, w is the soil mass moisture content.

5. The soil-water characteristic relationship prediction method based on big data analysis according to claim 4 is characterized in that: Combining equations (2), (3) and (4) yields the following relationship: （5）。 6. The soil-water characteristic relationship prediction method based on big data analysis according to claim 5 is characterized in that: Soil mass water content w With critical radius r The relationship between is the fractal model of soil pore structure: （6）。 7. The soil-water characteristic relationship prediction method based on big data analysis according to claim 6 is characterized in that: In the soil pore fractal model, the Young-Laplace equation is used to establish the capillary suction S Size and pore radius r The contact angle of the solid-liquid-gas three-phase interface θ Relationship: （7）； Where: r is the pore radius, S is the capillary suction force, T s is the surface tension of the liquid; Substituting equation (7) into equation (6) yields the capillary suction force S Soil moisture content w The relationship between （8）； The prediction model includes T s , θ , L , d s , D , ρ d , a total of 6 unknown parameters: T s is the surface tension, which is 0.073 N / m at 25°C; θ is the contact angle, which is equal to 0 during the dewetting process; d s is the relative density of the soil, measured by the pycnometer method; dry density ρ d Measured according to the oven-drying method.

8. The soil-water characteristic relationship prediction method based on big data analysis according to claim 1 is characterized in that: The specific method of step 2 includes: collecting the soil moisture content of the experimental test w ~Soil water suction S Curve data and its corresponding soil surface physical index parameters, including relative density, dry density, particle grading or others, and establish their database; soil pore fractal characteristic parameters include fractal dimension D and maximum aperture L The apparent physical indicators of soil include relative density, dry density, particle grading or others.

9. The soil-water characteristic relationship prediction method based on big data analysis according to claim 1 is characterized in that: The step 2 is to establish a corresponding relationship between soil pore fractal characteristic parameters and soil surface physical indicators based on a BP neural network, and the specific steps include: 2.1 According to formula (8), the left side of the formula As the vertical axis, the right side As the horizontal axis, draw a data point graph and perform a linear fit on the data points; 2.2 Obtaining the slope based on the linear fitting results and the intercept , calculate the fractal dimension D and maximum aperture L ; 2.3 For different dry densities in the database ρ d , particle size distribution d The soil samples were preprocessed, including cleaning, denoising, and normalization, and the data set was divided into a training set and a test set; 2.4 Determine the topological structure of the BP neural network, take the soil surface physical parameters as input, take the pore fractal parameters as output, and determine the number of hidden layer neurons according to the empirical formula; （9）； In the formula, n is the number of neurons in the hidden layer, m is the number of nodes in the input layer, l is the number of nodes in the output layer, k ∈(0,10); 2.5 Input the initial weights and thresholds into the neural network for training, calculate the error between the real data and the predicted data of the pore fractal parameters, and when the error value does not meet the preset accuracy requirement, feedback is given from the output layer to the input layer layer by layer, and the connection weights and thresholds between the layers are continuously adjusted until the error reaches the set stop condition; 2.6 Run the test set according to the optimized connection weights and thresholds to verify the prediction accuracy of the trained BP neural network.

Images