Neuron model prediction method for soil heavy metal form content

Abstract

The invention discloses a neuron model prediction method for soil heavy metal form content, and belongs to the technical field of soil environment monitoring and data mining. The method specifically comprises the following steps: taking key influence factors such as physicochemical properties (including total metal amount, pH value, cation exchange capacity CEC and the like) of a soil sample as model input variables, then constructing an artificial neural network prediction model, adjusting each parameter weight through an optimization algorithm, and establishing a quantitative relationship between the input variables and an output metal form; the model is verified and optimized, and high-precision prediction of the content of the heavy metal elements in the soil in different forms is achieved. The method can effectively overcome the defects that a traditional form analysis method is complex in operation and high in cost, and accurate prediction of the soil heavy metal form is achieved. The prediction method is an important way for expanding soil data information and reducing the detection workload, has important significance on soil data mining based on limited information, and is a quantitative prediction method with practical value.

Description

Technical Field

[0001] This invention belongs to the field of soil environmental monitoring and data mining technology, specifically involving a neural network model prediction method for the content of heavy metal speciation in soil, which can be used to quickly predict the content of heavy metal speciation in soil. Background Technology

[0002] With rapid industrialization and urbanization, soil heavy metal pollution has become an increasingly prominent problem. The environmental functions and effects of heavy metals are determined by their different valence states and existing forms. Traditional heavy metal speciation analysis methods, such as the Tessier sequential extraction method, while accurately determining the content of different heavy metal speciations, suffer from drawbacks such as cumbersome procedures, long processing times, and high costs, making them unsuitable for large-scale soil environmental surveys. Developing metal speciation prediction methods based on existing soil properties is an important way to expand soil data information and reduce the number of detection indicators, and is of great significance for soil data mining based on limited information.

[0003] Artificial neural networks (ANNs) exhibit unique advantages in predicting small sample data due to their powerful nonlinear modeling capabilities. However, existing models often neglect the transformation relationships between different speciations, resulting in prediction accuracy that fails to meet practical application requirements. In soil environmental monitoring practice, the national soil survey has accumulated a large amount of soil physicochemical property data, but it does not include data on the various speciations of heavy metals. How to predict the distribution of heavy metal speciation using existing conventional monitoring data has become a key technical problem urgently needing to be solved in the field of soil environmental research. Therefore, developing a heavy metal speciation prediction method based on a neural network model is of great significance for improving the soil environmental quality assessment system and guiding soil pollution prevention and control. Summary of the Invention

[0004] The purpose of this invention is to overcome the deficiencies in the prior art and provide a neural network model prediction method for the content of heavy metal speciation in soil. By integrating soil physicochemical properties and speciation data, a high-precision prediction model is constructed.

[0005] The specific technical solution adopted in this invention is as follows:

[0006] In a first aspect, the present invention provides a neural network model method for predicting the content of heavy metal speciation in soil, as detailed below:

[0007] S1: Using the physicochemical properties of soil samples from the target area and the content of different forms of heavy metals as a database, correlation analysis was used to screen out the physicochemical properties that are significantly related to the forms of heavy metals as input variables for the model.

[0008] S2: Construct an artificial neural network model, using the physicochemical property parameters obtained in S1 as input variables and the heavy metal speciation content as output variables, and optimize the prediction by assigning weights to the input variables;

[0009] S3: Verify the prediction accuracy of the artificial neural network model obtained in S2 and apply it to the content assessment of heavy metal speciation in the soil of the test area.

[0010] Preferably, in S1, the physicochemical properties include at least three of the following: total metal element content, pH value, soil cation exchange capacity, and soil organic matter.

[0011] Furthermore, the total content of the metal elements is one of the following: total manganese, total mercury, total cadmium, total lead, total copper, total chromium, total arsenic, total nickel, total iron, and total zinc.

[0012] Preferably, in S1, the content of different forms of heavy metal elements is obtained by the Tessier five-step continuous extraction method.

[0013] Preferably, in S1, the different forms of heavy metal elements include ion-exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic and sulfide-bound state, and residue state.

[0014] Preferably, in S2, the artificial neural network model employs multi-parameter nonlinear fitting.

[0015] Secondly, the present invention provides a neural network model prediction system for the content of heavy metal speciation in soil, comprising:

[0016] The input variable acquisition module is used to select physicochemical property parameters that are significantly related to heavy metal speciation as model input variables by using the physicochemical property parameters and the content of different forms of heavy metal elements in soil samples of the target area as a database and through correlation analysis.

[0017] The model building module is used to build an artificial neural network model. The physicochemical property parameters obtained by the input variable acquisition module are used as input variables and the content of heavy metal speciation is used as output variables. The prediction is achieved by optimizing the weights of the input variables.

[0018] The validation and application module is used to validate the prediction accuracy of the artificial neural network model obtained through the model building module and apply it to the content assessment of heavy metal speciation in the soil of the test area.

[0019] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the neural network model prediction method for soil heavy metal speciation as described in any of the first aspects.

[0020] Fourthly, the present invention provides a computer electronic device, including a memory and a processor;

[0021] The memory is used to store computer programs;

[0022] The processor is configured to, when executing the computer program, implement the neural network model prediction method for soil heavy metal speciation as described in any of the first aspects.

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] The prediction method provided by this invention is highly efficient, widely applicable, and has high prediction accuracy. By constructing an artificial neural network model, it fully considers the nonlinear relationship between heavy metal speciation and soil physicochemical properties, and the prediction root mean square error R... 2 The accuracy can reach above 0.85, with the error reduced to 16%, compared to the traditional single-parameter linear regression method (R²). 2 The model has significant advantages (typically below 0.5); it uses conventional soil survey data (such as pH, CEC, etc.) as input parameters, eliminating the need for complex speciation analysis and thus saving significant testing costs; the model is applicable to speciation prediction of various heavy metal elements and has strong versatility. Attached Figure Description

[0025] Figure 1 The basic principle of a neuron model (a) and a manganese speciation prediction network based on soil physicochemical properties (b) (where w represents the weight assignment).

[0026] Figure 2 The graphs show the relationship between predicted and measured values ​​for univariate, bivariate, and trivariate methods, using carbonate-bound manganese content as an example (a, with total manganese as a single input parameter; b, with total manganese and pH as dual input parameters; c, with total manganese, pH, and CEC as trivariate input parameters); (d) predicted regression coefficients (R²). 2 The relationship between the number of input parameters and the number of input parameters. In the figure, the dark shaded area represents the confidence interval, and the light shaded area represents the prediction band. Detailed Implementation

[0027] The present invention will be further described and illustrated below with reference to the accompanying drawings and specific embodiments. The technical features of each embodiment of the present invention can be combined accordingly, provided that there is no mutual conflict.

[0028] This invention provides a neural network model method for predicting the content of heavy metal speciation in soil. The specific steps of this prediction method are as follows:

[0029] S1: Using the physicochemical properties of soil samples from the target area and the content of different forms of heavy metals as a database, correlation analysis was used to screen out the physicochemical properties that are significantly related to the heavy metal forms as input variables for the model.

[0030] In a preferred embodiment of the present invention, the physicochemical properties parameters in this step include at least three of the following: total content of metal elements (such as total manganese, total mercury, total cadmium, total lead, total copper, total chromium, total arsenic, total nickel, total iron, total zinc, etc.), pH value, soil cation exchange capacity (CEC), and soil organic matter (SOM).

[0031] As a preferred embodiment of the present invention, the heavy metal elements in this step are detected by referring to the Tessier five-step continuous extraction method or other metal element speciation methods based on solvent (or solution) extraction principles, including ion-exchange state, carbonate-bound state, iron-manganese oxide-bound state, organic and sulfide-bound state, and residue state.

[0032] S2: Construct an artificial neural network model, using the physicochemical properties parameters obtained from S1 as input variables and the heavy metal speciation content as output variables, and optimize the prediction by assigning weights to the input variables.

[0033] In a preferred embodiment of the present invention, the artificial neural network model in this step employs multi-parameter nonlinear fitting.

[0034] In a preferred embodiment of the present invention, in this step, the prediction of the output parameter is achieved by adjusting the weights of the input parameters. Real farmland surface soil is used, and its physicochemical properties are measured to represent the essential parameters for soil surveys. Soil heavy metal speciation data obtained through the classic Tessier method are used as validation data for the model, correcting the weight (w) values, thereby achieving in-depth analysis of heavy metal speciation content.

[0035] S3: Verify the prediction accuracy of the artificial neural network model obtained in S2 and apply it to the content assessment of heavy metal speciation in the soil of the test area.

[0036] The methods and effects of the present invention will be specifically illustrated below through examples.

[0037] Example 1

[0038] This embodiment provides a neural network model prediction method for manganese speciation in farmland soil, such as... Figure 1 In a basic neuron model, the output parameters are predicted by adjusting the weights of the input parameters. The specific prediction method is as follows:

[0039] S1: Correlation analysis between soil physicochemical properties (including total manganese content, pH, organic matter content, and cation exchange capacity) and manganese speciation shows that total manganese content and pH are important factors affecting manganese speciation. When total manganese content is used as the single input parameter for linear fitting of manganese speciation, the calculation formula and results of the fitted equation are as follows:

[0040] (Y1 Y2 Y3 Y4) = (X1) × (w 11 w 21 w 31 w 41 )+(w 12 w 22 w 32 w 42 (1-1)

[0041]

[0042] Where X1 represents the total manganese content (g·kg) -1 ), w represents the weight value (dimensionless); Y1, Y2, Y3, and Y4 represent the contents of ion-exchangeable manganese, carbonate-bound manganese, iron-manganese oxide-bound manganese, and organic and sulfide-bound manganese, respectively (g·kg). -1 ); Let w represent the predicted regression coefficients for the four manganese forms. From equation (1-2), we can see that all weights w i1 (i=1,2,3,4) are all positive values. As shown in equation (1-3), the prediction effect of a single input parameter is relatively poor, and the regression coefficient (Ri) is... 2 The values ​​are generally below 0.5. Therefore, it is necessary to further increase the number of input parameters.

[0043] S2: Using total manganese content and pH as dual input parameters, a linear fit was performed on the manganese speciation. The resulting fitting equation and results are as follows:

[0044]

[0045] From equation (2-2), we can see that all weights w i1 (i = 1, 2, 3, 4) are all positive values, indicating that the content of manganese in each form is positively correlated with the total manganese content; while all weights w i2 (i = 1, 2, 3, 4) are all negative, indicating a negative correlation between the total pH and the content of manganese in each form. Comparing equations (1-3) and (2-3), it can be seen that the prediction effect of dual-input parameters is significantly improved compared to that of single-input parameters, especially the regression coefficient (R²) for the prediction of manganese content in ion-exchange states. 2 The prediction accuracy improved rapidly from 0.19 to 0.85, and the prediction accuracy for other manganese forms also improved, but the regression coefficient remained around 0.5.

[0046] S3: When further increasing soil SOM content and CEC as input parameters, the following fitting equation calculation formula and results can be obtained:

[0047]

[0048] Comparing equations (3-3) and (2-3), it can be seen that the prediction effect of four input parameters is significantly improved compared with the prediction effect of two input parameters. In particular, the prediction effect of carbonate-bound manganese content, iron-manganese oxide-bound manganese content, and organic and sulfide-bound manganese content is significantly improved, with the regression coefficient increasing from 0.5 to about 0.7. However, the improvement in the prediction effect of ion-exchange state is limited, only increasing from 0.85 to 0.86.

[0049] Figure 2 This shows the relationship between the predicted and measured values ​​of manganese content in the carbonate-bound state in this example. Figure 2 The values ​​ac represent the prediction performance with total manganese as a single input parameter, total manganese and pH as dual input parameters, and total manganese, pH, and CEC as trivariate input parameters, respectively. As the number of input parameters increases, the prediction performance significantly improves, and the prediction interval (light shading) narrows considerably. For data within the confidence interval, the maximum prediction error decreases from 65.66% for the univariate model to 16.52% for the trivariate model, demonstrating good predictive performance. Figure 2 d shows the predicted regression coefficients (R²). 2 The relationship between the number of input variables and the number of input parameters. It is particularly important to note that when the number of input variables increases from 3 to 4, the fitting effect does not improve further (R²). 2 =0.78), indicating that the multivariate fitting exhibits overfitting after introducing the SOM variable. Overfitting leads to meaningless fluctuations in the fitted training samples, but the predictive ability (generalization ability) for real data decreases. Therefore, in the dataset of this embodiment, the multivariate fitting model constructed with pH, ​​total manganese content, and CEC as prediction parameters is: Y = 0.01798X1 - 0.01292X2 + 0.00192X3 + 0.07426 (R = 0.78). 2 =0.7888) has the best predictive performance.

[0050] This invention uses key influencing factors such as the physicochemical properties of soil samples (including total metal content, pH value, cation exchange capacity (CEC), etc.) as model input variables. An artificial neural network prediction model is then constructed, and the weights of each parameter are adjusted through optimization algorithms to establish a quantitative relationship between the input variables and the output metal speciation. The model is validated and optimized to achieve high-precision prediction of the content of different speciations of heavy metals in soil. This method effectively overcomes the drawbacks of traditional speciation analysis methods, such as complexity and high cost, and achieves accurate prediction of heavy metal speciation in soil. This prediction method is an important way to expand soil data information and reduce the workload of detection, and it is of great significance for soil data mining based on limited information. It is a practically valuable quantitative prediction method.

[0051] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A neuron-based method for predicting the content of heavy metal speciation in soil, characterized in that, Specifically as follows: S1: Using the physicochemical properties of soil samples from the target area and the content of different forms of heavy metals as a database, correlation analysis was used to screen out the physicochemical properties that are significantly related to the forms of heavy metals as input variables for the model. S2: Construct an artificial neural network model, using the physicochemical property parameters obtained in S1 as input variables and the heavy metal speciation content as output variables, and optimize the prediction by assigning weights to the input variables; S3: Verify the prediction accuracy of the artificial neural network model obtained in S2 and apply it to the content assessment of heavy metal speciation in the soil of the test area.

2. The method for predicting the content of heavy metal speciation in soil using a neuron model according to claim 1, characterized in that, In S1, the physicochemical properties include at least three of the following: total metal element content, pH value, soil cation exchange capacity, and soil organic matter.

3. The method for predicting soil heavy metal speciation using a neuron model according to claim 2, characterized in that, The total content of the metal elements is one of the following: total manganese, total mercury, total cadmium, total lead, total copper, total chromium, total arsenic, total nickel, total iron, and total zinc.

4. The method for predicting the content of heavy metal speciation in soil using a neuron model according to claim 1, characterized in that, In S1, the contents of different forms of heavy metal elements are obtained by the Tessier five-step continuous extraction method.

5. The method for predicting the content of heavy metal speciation in soil using a neuron model according to claim 1, characterized in that, In S1, the different forms of heavy metal elements include ion-exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic and sulfide-bound state, and residue state.

6. The method for predicting the content of heavy metal speciation in soil using a neuron model according to claim 1, characterized in that, In S2, the artificial neural network model employs multi-parameter nonlinear fitting.

7. A neural network model prediction system for heavy metal speciation in soil, characterized in that, include: The input variable acquisition module is used to select physicochemical property parameters that are significantly related to heavy metal speciation as model input variables by using the physicochemical property parameters and the content of different forms of heavy metal elements in soil samples of the target area as a database and through correlation analysis. The model building module is used to build an artificial neural network model. The physicochemical property parameters obtained by the input variable acquisition module are used as input variables and the content of heavy metal speciation is used as output variables. The prediction is achieved by optimizing the weights of the input variables. The validation and application module is used to validate the prediction accuracy of the artificial neural network model obtained through the model building module and apply it to the content assessment of heavy metal speciation in the soil of the test area.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the neuron model prediction method for soil heavy metal speciation as described in any one of claims 1 to 6.

9. A computer electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to, when executing the computer program, implement the neural network model prediction method for soil heavy metal speciation as described in any one of claims 1 to 6.

Images