Risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning

Abstract

The present invention discloses a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning, including: collecting soil data and leaching data of heavy metal pollution mining areas, wherein the soil data includes pollution data and remediation data; comparing pollution data and remediation data to obtain impact data, dividing the impact data based on the number of eluents, and obtaining mixed impact and independent impact; performing impact analysis on independent impacts according to leaching data to obtain independent impact strength, and performing impact analysis on mixed impacts according to independent impact strength to obtain mixed impact strength; obtaining a remediation risk coefficient according to the independent impact strength and the mixed impact strength, and establishing a remediation risk assessment model based on the remediation risk coefficient; optimizing the remediation risk assessment model, inputting the data to be assessed into the remediation risk assessment model, and outputting the assessment result. The method can quickly and accurately assess the risk of leaching and remediation of heavy metal pollution in mining areas, and provides a scientific basis for pollution control.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental pollution control, and in particular to a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning. Background Art

[0002] Mining activities are often accompanied by heavy metal pollution problems. These heavy metals, such as lead, mercury, and cadmium, are seriously harmful to the environment and human health. Traditional heavy metal pollution assessment methods mainly rely on chemical analysis and biological detection techniques. Although these methods can provide quantitative data on pollution levels, they have many limitations. First, these methods usually require complex laboratory operations and long sample processing, and cannot achieve rapid assessment. Secondly, existing technologies often lack the ability to monitor the dynamic changes of risks in the process of pollution remediation in real time. In addition, traditional assessment methods have deficiencies in real-time updating and processing of data, and are difficult to adapt to changing site conditions and remediation needs.

[0003] The present invention aims to solve the above problems and proposes a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning. The method can monitor pollution data in real time, quickly assess remediation risks, and provide scientific and accurate decision support for pollution control. Summary of the invention

[0004] The purpose of the present invention is to provide a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Collect soil data and leaching data of heavy metal polluted mining areas, wherein the soil data includes pollution data and remediation data;

[0008] Comparing the contamination data with the repair data to obtain impact data, and dividing the impact data based on the amount of eluent to obtain mixed impact and independent impact;

[0009] Performing an impact analysis on the independent impact according to the elution data to obtain an independent impact strength, and performing an impact analysis on the mixed impact according to the independent impact strength to obtain a mixed impact strength;

[0010] Obtaining a restoration risk coefficient according to the independent impact intensity and the mixed impact intensity, and establishing a restoration risk assessment model based on the restoration risk coefficient;

[0011] The repair risk assessment model is optimized, the data to be assessed is input into the repair risk assessment model, and an assessment result is output.

[0012] Further, a method for collecting soil data and leaching data of a heavy metal contaminated mining area, wherein the soil data includes contamination data and remediation data, comprises:

[0013] The pollution data refers to the soil data before leaching and remediation, and the remediation data refers to the soil data after leaching and remediation. The leaching data includes the concentration, leaching time and type of the elution agent used in the elution and remediation.

[0014] Further, the method of comparing the pollution data and the repair data to obtain impact data, and dividing the impact data based on the amount of eluent to obtain mixed impact and independent impact includes:

[0015] Calculate the change ratio and change data of the repair data relative to the contaminated data,

[0016] Using statistical analysis, we screened out the features that have a significant impact on the restoration risk assessment:

[0017] Mutual Information:

[0018]

[0019] Where ML(X,Y) is the mutual information between feature set X and feature set Y, Y is the influence data, X is the elution data, x and y are the indexes of the features, p(x,y) is the joint probability distribution of x and y, p(x) is the marginal probability distribution of X, p(y) is the marginal probability distribution of Y, and f impact (x,y) is the weight function between x and y based on prior knowledge,

[0020] Mutual information adjusted by complexity:

[0021] MI complex (X,Y)=ML(X,Y)·g complexity (X,Y)

[0022] Among them, MI complex (X,Y) is the complexity-adjusted mutual information between feature set X and feature set Y, g complexity (X,Y) is the complexity function,

[0023] Feature Importance Score:

[0024]

[0025] in, and is the weight coefficient, t w is the change ratio, Δx noisy represents the noise effect of X,

[0026] Remove the change ratios whose feature importance scores are less than the importance threshold, and use the remaining change ratios as impact data.

[0027] The impact data corresponding to the elution remediation using only one eluent are classified as independent impacts, and the remaining impact data are classified as mixed impacts.

[0028] Further, the method of performing an impact analysis on the independent impact according to the elution data to obtain the independent impact strength includes:

[0029] The type of eluent in the elution data is transformed into a numerical vector using unique hot encoding. The elution data and contamination data are used as influencing features, and the mapping relationship between the influencing features of independent influences and the influencing data is learned using a generative adversarial network:

[0030] The loss function of the discriminator and generator in the generative adversarial network is:

[0031]

[0032] Among them, L(J,K) is the adversarial loss function of the discriminator and the generator, J is the discriminator, K is the generator, and a is the sample data received by J. is the weight of the sample data, is the weight of the generated sample generated by the generator, J(a) is the probability that a is the sample data output by J, z is the random noise vector, q real (a) is the distribution of sample data, q z (z) is the distribution of z, E is the expectation, λ and μ are regularization coefficients, w is the network weight, and the gradient calculation formula of the discriminator is:

[0033]

[0034] in, is the gradient of the discriminator's parameters, the gradient calculation formula of the generator is:

[0035]

[0036] in, is the gradient of the parameters of the generator, and the parameters are adjusted by the gradient of the parameters to minimize the loss function.

[0037] For the missing influence features of a single influence in the influence features of mixed influence, the influence data of the missing influence features are generated through the trained generative adversarial network.

[0038] Based on prior knowledge, the output of the Gaussian mixture regression model is divided into positive impact and negative impact, and the independent impact intensity is obtained.

[0039] Furthermore, the method of performing an impact analysis on the mixed impact according to the independent impact strength to obtain the mixed impact strength includes:

[0040] According to the pollution data and leaching data of mixed impact, the corresponding independent impact intensity is obtained as the feature vector, and the impact data of mixed impact is used as the target vector. The neural network algorithm is used to establish the mixed impact intensity prediction model:

[0041] The expected function of the prediction result is:

[0042]

[0043] Among them, f(x,y) is the joint probability density function of x and y, x is the feature vector, y is the target vector,

[0044] The joint probability density function is obtained by nonparametric estimation:

[0045]

[0046] Among them, i is the index of the feature vector and the target vector, n is the number of i, k is the dimension of x, σ is the standard deviation, T represents the transpose of the matrix, and w is the weight vector.

[0047] Prediction results:

[0048]

[0049] The leaching data and pollution data are arranged and combined, and the new mixed impact data are predicted by the mixed impact intensity prediction model.

[0050] Based on prior knowledge, the output of the mixed impact intensity prediction model is divided into positive impact and negative impact to obtain the mixed impact intensity.

[0051] Furthermore, the method for obtaining the restoration risk coefficient according to the independent impact strength and the mixed impact strength includes:

[0052] The calculation formula of repair risk factor is:

[0053]

[0054] Where M is the number of negative impacts, N is the number of positive impacts, L is the impact intensity of leaching and repair, is the independent impact intensity or the mixed impact intensity, m is the index of negative impact, and f impact is the weight of the mth negative impact, P is the total number of negative impacts of the eluent used in the elution repair, and Q is the total number of positive impacts of the eluent used in the elution repair.

[0055] Furthermore, a method for establishing a restoration risk assessment model based on the restoration risk coefficient includes:

[0056] A data set was established using pollution data, leaching data and remediation risk coefficients, and the data set was divided into a training set and a test set. The random forest algorithm was used to learn the mapping pattern between pollution data, leaching data and remediation risk coefficients through the training set, and a risk assessment model was established. The test set was used to optimize the parameters of the risk assessment model.

[0057] Furthermore, the method for optimizing the repair risk assessment model includes:

[0058] The cuckoo search algorithm is used to optimize the model parameters of the risk assessment model:

[0059] Initialize the number and location of bird nests. Each nest location contains a set of model parameters. The fitness function is set to the mean square error of the risk assessment model.

[0060] Perform an initial fitness evaluation on each bird's nest, perform fitness repair according to the Levy flight search mechanism, set the maximum number of searches to N, and update the nest position:

[0061] η i,t+1 =η i,t +l·(L+∈)+λ·(η i,t -η mean ),

[0062]

[0063] where η i,t is the nest position of the i-th bird's nest repaired for the tth time, 1 is the step length, L is the correction factor, u and v are uniform distribution functions, i * is the nest location with the highest fitness, ∈ is a random disturbance, η mean is the average location of all bird nests,

[0064] Uniform distribution function:

[0065]

[0066] Where Γ is the gamma function, λ is the adjustment parameter, the value range is [0, 1], λ0 is the initial adjustment parameter, Δλ is the change step of the adjustment parameter,

[0067] Formula for repairing a bird's nest:

[0068]

[0069] where η j,ι For and η i,t The neighboring bird's nest, is the probability of repairing the bird's nest, with a value range of [0, 1],

[0070] According to the greedy rule, the bird nests with higher fitness after repair are retained and compared with the bird nests with the highest current fitness until the maximum number of searches is reached, then the search is stopped and the bird nest with the highest fitness is selected.

[0071] The beneficial effects of the present invention are:

[0072] The present invention is a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning. Compared with the prior art, the present invention has the following technical effects:

[0073] The use of machine learning technology has improved the accuracy and real-time performance of risk assessment.

[0074] Impact analysis enables more precise identification and quantification of key risk factors in the pollution remediation process.

[0075] The application of optimization algorithms further improves the predictive performance and robustness of the model.

[0076] This method can be widely used in different types of heavy metal pollution remediation projects in mining areas and has good universality and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a flow chart of the steps of a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning in the present invention. DETAILED DESCRIPTION

[0078] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0079] The present invention provides a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning, comprising the following steps:

[0080] like Figure 1 As shown, in this embodiment, the following steps are included:

[0081] Collect soil data and leaching data of heavy metal polluted mining areas, wherein the soil data includes pollution data and remediation data;

[0082] In the actual assessment, soil data includes the heavy metal content, microbial biomass and pH value of the soil; the collected soil data are:

[0083]

[0084] In the table, pollution indicates pollution data, and repair indicates repair data; the elution data are:

[0085] Sample No. Eluent Type Eluent concentration (g / L) Washing time (h) A Eluent 1 0.5 2 B Eluent 1 and 2 0.7 3 C Eluent 2 0.5 2 D Eluent 1 0.6 4 ;

[0086] Comparing the contamination data with the repair data to obtain impact data, and dividing the impact data based on the amount of eluent to obtain mixed impact and independent impact;

[0087] In the actual evaluation, the impact data are:

[0088]

[0089] In the table, “removed” indicates the proportion of changes that were removed, sample B is a mixed influence, and the remaining samples are independent influences;

[0090] Performing an impact analysis on the independent impact according to the elution data to obtain an independent impact strength, and performing an impact analysis on the mixed impact according to the independent impact strength to obtain a mixed impact strength;

[0091] In the actual evaluation, the independent impact strength includes: the impact data of samples A, C and D, and the predicted impact data of sample B using eluent 1 and eluent 2 alone: ​​[18.18, 28.57, 16.7, 4.00], [25, 36.5, 29, -4.5]; the mixed impact strength includes the impact data of sample B, and the predicted impact data of samples A, C and D using a mixture of eluent 1 and eluent 2: [21.59, 30.59, 28.33, 4.40], [19.59, 29.59, 25.33, -4.00], [17.59, 28.59, 23.33, -8.00], among which the increase in the proportion of microbial biomass change is a positive impact, and the decrease in the proportion of heavy metal change is a positive impact;

[0092] Obtaining a restoration risk coefficient according to the independent impact intensity and the mixed impact intensity, and establishing a restoration risk assessment model based on the restoration risk coefficient;

[0093] In the actual assessment, the repair risk coefficients are: [0.18, 0.1, 0.14] for sample A using eluents 1, 2, and mixed eluents; [0.16, 0.08, 0.12] for sample B using eluents 1, 2, and mixed eluents; [0.14, 0.06, 0.10] for sample C using eluents 1, 2, and mixed eluents; and [0.12, 0.04, 0.08] for sample D using eluents 1, 2, and mixed eluents.

[0094] The repair risk assessment model is optimized, the data to be assessed is input into the repair risk assessment model, and an assessment result is output.

[0095] In this embodiment, a method for collecting soil data and leaching data of a heavy metal polluted mining area, wherein the soil data includes pollution data and remediation data, includes:

[0096] The pollution data refers to the soil data before leaching and remediation, and the remediation data refers to the soil data after leaching and remediation. The leaching data includes the concentration, leaching time and type of the elution agent used in the elution and remediation.

[0097] In this embodiment, the method of comparing the contamination data and the repair data to obtain the impact data, and dividing the impact data based on the amount of the eluent to obtain the mixed impact and the independent impact includes:

[0098] Calculate the change ratio and change data of the repair data relative to the contaminated data,

[0099] Using statistical analysis, we screened out the features that have a significant impact on the restoration risk assessment:

[0100] Mutual Information:

[0101]

[0102] Where ML(X,Y) is the mutual information between feature set X and feature set Y, Y is the influence data, X is the elution data, x and y are the indexes of the features, p(x,y) is the joint probability distribution of x and y, p(x) is the marginal probability distribution of X, p(y) is the marginal probability distribution of Y, and f impact (x,y) is the weight function between x and y based on prior knowledge,

[0103] Mutual information adjusted by complexity:

[0104] MI complex (X,Y)=ML(X,Y)·g complexity (X,Y)

[0105] Among them, MI complex (X,Y) is the complexity-adjusted mutual information between feature set X and feature set Y, g complexity (X,Y) is the complexity function,

[0106] Feature Importance Score:

[0107]

[0108] in, and is the weight coefficient, t w is the change ratio, Δx noisy represents the noise effect of X,

[0109] Remove the change ratios whose feature importance scores are less than the importance threshold, and use the remaining change ratios as impact data.

[0110] The impact data corresponding to the elution remediation using only one eluent are classified as independent impacts, and the remaining impact data are classified as mixed impacts.

[0111] In this embodiment, the method of performing impact analysis on the independent impact according to the elution data to obtain the independent impact strength includes:

[0112] The type of eluent in the elution data is transformed into a numerical vector using unique hot encoding. The elution data and contamination data are used as influencing features, and the mapping relationship between the influencing features of independent influences and the influencing data is learned using a generative adversarial network:

[0113] The loss function of the discriminator and generator in the generative adversarial network is:

[0114]

[0115] Among them, L(J,K) is the adversarial loss function of the discriminator and the generator, J is the discriminator, K is the generator, and a is the sample data received by J. is the weight of the sample data, is the weight of the generated sample generated by the generator, J(a) is the probability that a is the sample data output by J, z is the random noise vector, q real (a) is the distribution of sample data, q z (z) is the distribution of z, E is the expectation, λ and μ are regularization coefficients, w is the network weight, and the gradient calculation formula of the discriminator is:

[0116]

[0117] in, is the gradient of the discriminator's parameters, the gradient calculation formula of the generator is:

[0118]

[0119] in, is the gradient of the parameters of the generator, and the parameters are adjusted by the gradient of the parameters to minimize the loss function.

[0120] For the missing influence features of a single influence in the influence features of mixed influence, the influence data of the missing influence features are generated through the trained generative adversarial network.

[0121] Based on prior knowledge, the output of the Gaussian mixture regression model is divided into positive impact and negative impact, and the independent impact intensity is obtained.

[0122] In this embodiment, the method of performing impact analysis on the mixed impact according to the independent impact strength to obtain the mixed impact strength includes:

[0123] According to the pollution data and leaching data of mixed impact, the corresponding independent impact intensity is obtained as the feature vector, and the impact data of mixed impact is used as the target vector. The neural network algorithm is used to establish the mixed impact intensity prediction model:

[0124] The expected function of the prediction result is:

[0125]

[0126] Among them, f(x,y) is the joint probability density function of x and y, x is the feature vector, y is the target vector,

[0127] The joint probability density function is obtained by nonparametric estimation:

[0128]

[0129] Among them, i is the index of the feature vector and the target vector, n is the number of i, k is the dimension of x, σ is the standard deviation, T represents the transpose of the matrix, and w is the weight vector.

[0130] Prediction results:

[0131]

[0132] The leaching data and pollution data are arranged and combined, and the new mixed impact data are predicted by the mixed impact intensity prediction model.

[0133] Based on prior knowledge, the output of the mixed impact intensity prediction model is divided into positive impact and negative impact to obtain the mixed impact intensity.

[0134] In this embodiment, the method for obtaining the restoration risk coefficient according to the independent impact strength and the mixed impact strength includes:

[0135] The calculation formula of repair risk factor is:

[0136]

[0137] Where M is the number of negative impacts, N is the number of positive impacts, L is the impact intensity of leaching and repair, is the independent impact intensity or the mixed impact intensity, m is the index of negative impact, and f impact is the weight of the mth negative impact, P is the total number of negative impacts of the eluent used in the elution repair, and Q is the total number of positive impacts of the eluent used in the elution repair.

[0138] In this embodiment, the method for establishing a restoration risk assessment model based on the restoration risk coefficient includes:

[0139] A data set was established using pollution data, leaching data and remediation risk coefficients, and the data set was divided into a training set and a test set. The random forest algorithm was used to learn the mapping pattern between pollution data, leaching data and remediation risk coefficients through the training set, and a risk assessment model was established. The test set was used to optimize the parameters of the risk assessment model.

[0140] In this embodiment, the method for optimizing the repair risk assessment model includes:

[0141] The cuckoo search algorithm is used to optimize the model parameters of the risk assessment model:

[0142] Initialize the number and location of bird nests. Each nest location contains a set of model parameters. The fitness function is set to the mean square error of the risk assessment model.

[0143] Perform an initial fitness evaluation on each bird's nest, perform fitness repair according to the Levy flight search mechanism, set the maximum number of searches to N, and update the nest position:

[0144] η i,t+1 =η i,t +l·(L+∈)+λ·(η i,t -η mean ),

[0145]

[0146] where η i,t is the nest position of the i-th bird's nest repaired for the tth time, 1 is the step length, L is the correction factor, u and v are uniform distribution functions, i * is the nest location with the highest fitness, ∈ is a random disturbance, η mean is the average location of all bird nests,

[0147] Uniform distribution function:

[0148]

[0149] Where Γ is the gamma function, λ is the adjustment parameter, the value range is [0, 1], λ0 is the initial adjustment parameter, Δλ is the change step of the adjustment parameter,

[0150] Formula for repairing a bird's nest:

[0151]

[0152] where η j,ι For and η i,t The neighboring bird's nest, is the probability of repairing the bird's nest, with a value range of [0, 1],

[0153] According to the greedy rule, the bird nests with higher fitness after repair are retained and compared with the bird nests with the highest current fitness until the maximum number of searches is reached, then the search is stopped and the bird nest with the highest fitness is selected.

[0154] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning, characterized in that: The following steps are involved: Collect soil data and leaching data of heavy metal polluted mining areas, wherein the soil data includes pollution data and remediation data; Comparing the contamination data with the repair data to obtain impact data, and dividing the impact data based on the amount of eluent to obtain mixed impact and independent impact; Performing an impact analysis on the independent impact according to the elution data to obtain an independent impact strength, and performing an impact analysis on the mixed impact according to the independent impact strength to obtain a mixed impact strength; Obtaining a restoration risk coefficient according to the independent impact intensity and the mixed impact intensity, and establishing a restoration risk assessment model based on the restoration risk coefficient; Optimizing the restoration risk assessment model, inputting the data to be assessed into the restoration risk assessment model, and outputting the assessment result; The method of comparing the contamination data and the repair data to obtain impact data, and dividing the impact data based on the amount of eluent to obtain mixed impact and independent impact includes: Calculate the change ratio and change data of the repair data relative to the contaminated data, Using statistical analysis, we screened out the features that have a significant impact on the restoration risk assessment: Mutual Information: Where ML(X,Y) is the mutual information between feature set X and feature set Y, Y is the influence data, X is the elution data, x and y are the indexes of the features, p(x,y) is the joint probability distribution of x and y, p(x) is the marginal probability distribution of X, p(y) is the marginal probability distribution of Y, and f impact (x,y) is the weight function between x and y based on prior knowledge, Mutual information adjusted by complexity: MY complex (X,Y)=ML(X,Y)·g complexity (X,Y) Among them, MI complex (X,Y) is the complexity-adjusted mutual information between feature set X and feature set Y, g complexity (X,Y) is the complexity function, Feature Importance Score: in, and is the weight coefficient, t w is the change ratio, Δx noisy represents the noise effect of X, Remove the change ratios whose feature importance scores are less than the importance threshold, and use the remaining change ratios as impact data. The impact data corresponding to the elution remediation using only one eluent are classified as independent impacts, and the remaining impact data are classified as mixed impacts; The method of performing an impact analysis on the independent impact according to the elution data to obtain the independent impact strength includes: The type of eluent in the elution data is transformed into a numerical vector using unique hot encoding. The elution data and contamination data are used as influencing features, and the mapping relationship between the influencing features of independent influences and the influencing data is learned using a generative adversarial network: The loss function of the discriminator and generator in the generative adversarial network is: Among them, L(J,K) is the adversarial loss function of the discriminator and the generator, J is the discriminator, K is the generator, and a is the sample data received by J. is the weight of the sample data, is the weight of the generated sample generated by the generator, J(a) is the probability that a output by J is the sample data, is a random noise vector, q real (a) is the distribution of sample data, for distribution, E is the expectation, λ and μ are regularization coefficients, w is the network weight, and the gradient calculation formula of the discriminator is: in, is the gradient of the discriminator's parameters, the gradient calculation formula of the generator is: in, is the gradient of the parameters of the generator, and the parameters are adjusted by the gradient of the parameters to minimize the loss function. For the missing influence features of a single influence in the influence features of mixed influence, the influence data of the missing influence features are generated through the trained generative adversarial network. Based on prior knowledge, the output of the Gaussian mixture regression model is divided into positive impact and negative impact, and the independent impact strength is obtained; The method of performing an impact analysis on the mixed impact according to the independent impact strength to obtain the mixed impact strength includes: According to the pollution data and leaching data of mixed impact, the corresponding independent impact intensity is obtained as the feature vector, and the impact data of mixed impact is used as the target vector. The neural network algorithm is used to establish the mixed impact intensity prediction model: The expected function of the prediction result is: Among them, f(x,y) is the joint probability density function of x and y, x is the feature vector, y is the target vector, The joint probability density function is obtained by nonparametric estimation: Among them, i is the index of the feature vector and the target vector, n is the number of i, k is the dimension of x, σ is the standard deviation, T represents the transpose of the matrix, and w is the weight vector. Prediction results: The leaching data and pollution data are arranged and combined, and the new mixed impact data are predicted by the mixed impact intensity prediction model. Based on prior knowledge, the output of the mixed impact intensity prediction model is divided into positive impact and negative impact to obtain the mixed impact intensity; The method for obtaining the restoration risk coefficient according to the independent impact intensity and the mixed impact intensity includes: The calculation formula of repair risk factor is: Where M is the number of negative impacts, N is the number of positive impacts, L is the impact intensity of leaching and repair, is the independent impact intensity or the mixed impact intensity, m is the index of negative impact, and f impact is the weight of the mth negative impact, P is the total number of negative impacts of the eluent used in the elution repair, and Q is the total number of positive impacts of the eluent used in the elution repair.

2. According to claim 1, a risk assessment method for leaching and remediation of heavy metal pollution in mining areas based on machine learning is characterized in that: A method for collecting soil data and leaching data of a heavy metal contaminated mining area, wherein the soil data includes contamination data and remediation data, comprises: The pollution data refers to the soil data before leaching and remediation, and the remediation data refers to the soil data after leaching and remediation. The leaching data includes the concentration, leaching time and type of the elution agent used in the elution and remediation.

3. According to the risk assessment method of heavy metal pollution leaching and remediation in mining areas based on machine learning in claim 1, it is characterized in that: The method for establishing a restoration risk assessment model based on the restoration risk coefficient comprises: A data set was established using pollution data, leaching data and remediation risk coefficients, and the data set was divided into a training set and a test set. The random forest algorithm was used to learn the mapping pattern between pollution data, leaching data and remediation risk coefficients through the training set, and a risk assessment model was established. The test set was used to optimize the parameters of the risk assessment model.

4. According to the risk assessment method of heavy metal pollution leaching and remediation in mining areas based on machine learning in claim 1, it is characterized in that: The method for optimizing the restoration risk assessment model comprises: The cuckoo search algorithm is used to optimize the model parameters of the risk assessment model: Initialize the number and location of bird nests. Each nest location contains a set of model parameters. The fitness function is set to the mean square error of the risk assessment model. Perform an initial fitness evaluation on each bird's nest, perform fitness repair according to the Levy flight search mechanism, set the maximum number of searches to N, and update the nest position: or i,t+1 =the i,t +l·(L+∈)+λ·(η i,t -or mean ), where η i,t is the nest position of the i-th bird's nest repaired for the tth time, l is the step length, L is the correction factor, u and v are uniform distribution functions, i * is the nest location with the highest fitness, ∈ is a random disturbance, η mean is the average location of all bird nests, Uniform distribution function: Where Γ is the gamma function, λ is the adjustment parameter, the value range is [0, 1], λ0 is the initial adjustment parameter, Δλ is the change step of the adjustment parameter, Formula for repairing a bird's nest: where η j,ι For and η i,t The neighboring bird's nest, is the probability of repairing the bird's nest, with a value range of [0, 1], According to the greedy rule, the bird nests with higher fitness after repair are retained and compared with the bird nests with the highest current fitness until the maximum number of searches is reached, then the search is stopped and the bird nest with the highest fitness is selected.

Images