Iron ore-inorganic ion combination stability evaluation method based on machine learning

Abstract

This invention belongs to the field of environmental geochemistry and data-driven modeling technology, specifically relating to a machine learning-based method for evaluating the binding stability of iron minerals and inorganic ions. The method includes the collection, organization, and preprocessing of experimental data on the adsorption and desorption of iron minerals and inorganic ions; the construction and performance evaluation of adsorption and desorption rate prediction models; and the evaluation of the binding stability of iron minerals and inorganic ions based on the prediction results. This invention systematically acquires iron mineral-inorganic ion adsorption and desorption data reported in the literature, employs various mainstream machine learning algorithms for modeling, compares model performance, and selects the optimal adsorption and desorption rate prediction model. Based on the optimal prediction model, the binding stability of different iron mineral-inorganic ion systems can be quantitatively evaluated, thereby providing methodological support for environmental risk assessment of inorganic pollutants and the formulation of pollution prevention and remediation strategies.

Description

Technical Field

[0001] This invention belongs to the field of environmental geochemistry and data-driven modeling technology, specifically involving a machine learning-based method for evaluating the stability of iron mineral-inorganic ion binding. Background Technology

[0002] Iron minerals are widely distributed in environmental media such as soil, water bodies, and sediments. The adsorption-desorption process between iron minerals and inorganic ions is a crucial interfacial reaction influencing the environmental fate of inorganic ions. The binding stability of inorganic ions on the surface of iron minerals essentially reflects their fixation capacity during adsorption and their release tendency during desorption. Therefore, accurately evaluating the binding stability of the iron mineral-inorganic ion system based on the entire adsorption-desorption process is of great significance for the study of the environmental behavior of inorganic pollutants, risk identification, and pollution control.

[0003] In existing technologies, research on the interfacial interaction between iron minerals and inorganic ions mainly relies on batch adsorption experiments, desorption experiments, and methods such as adsorption thermodynamic models, kinetic models, and surface complexation models. These methods are primarily used to study the adsorption and desorption behavior of inorganic ions on iron mineral surfaces and the interfacial interaction mechanisms, but they still fall short of meeting the practical needs for quantitative evaluation of bonding stability. Specifically, traditional experimental methods have long testing cycles and low throughput, making it difficult to meet the rapid evaluation requirements of different iron mineral-inorganic ion systems and under different environmental conditions. Furthermore, thermodynamic models, kinetic models, and surface complexation models typically require specific assumptions and their parameter acquisition is complex, thus limiting their applicability and predictive ability across different iron mineral types, inorganic ion types, and environmental conditions.

[0004] In recent years, machine learning methods have been gradually applied to the field of predicting the environmental behavior of pollutants, providing new technical approaches for rapid predictive analysis in complex systems. However, existing machine learning technologies mainly focus on predicting adsorption behavior, lacking effective prediction of desorption behavior. Therefore, it is not yet possible to quantitatively evaluate the binding stability of iron mineral-inorganic ion systems based on information from the entire adsorption-desorption process. Thus, it is necessary to provide a machine learning-based method for evaluating the binding stability of iron mineral-inorganic ion systems, enabling rapid quantitative evaluation of the binding stability of different iron mineral-inorganic ion systems, and providing methodological support for environmental risk assessment of inorganic pollutants and the formulation of pollution prevention and remediation strategies. Summary of the Invention

[0005] To address the problems of high experimental costs, difficulty in simultaneously considering adsorption and desorption behaviors, insufficient cross-system prediction capabilities, and inaccurate evaluation of binding capacity in existing technologies, the present invention aims to provide a machine learning-based method for evaluating the binding stability of iron minerals and inorganic ions.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A machine learning-based method for evaluating the stability of iron mineral-inorganic ion binding includes the following steps:

[0008] S1. Establish a dataset: Collect data on the adsorption and desorption rates of iron minerals and inorganic ions, as well as related experimental conditions, properties of iron minerals and inorganic ions, and establish a dataset for machine learning prediction models;

[0009] S2. Data preprocessing: Clean the established dataset, remove duplicate data and fill in missing values, and then standardize the cleaned data;

[0010] S3. Feature Filtering: Perform correlation and importance analysis on the variables in the dataset, filter the input features, remove features with strong correlation and low importance, and achieve feature dimensionality reduction;

[0011] S4. Model building: The dataset processed in step S3 is used to predict the adsorption rate and desorption rate of iron mineral-inorganic ions using different machine learning prediction models. The prediction performance of each model is evaluated, and the best prediction model for adsorption rate and desorption rate is selected respectively.

[0012] S5. Combination stability evaluation: Based on the optimal prediction model of adsorption rate and desorption rate constructed in step S4, the target iron mineral-inorganic ion system is predicted, and the binding stability of the target iron mineral-inorganic ion system is evaluated using the adsorption-desorption retention amount RM.

[0013] The flowchart of the above-mentioned machine learning-based method for evaluating the stability of iron mineral-inorganic ion binding is shown below. Figure 1 As shown.

[0014] Further, in step S1, the iron minerals include iron oxides or iron hydroxyl oxides (such as ferrite, goethite, hematite, etc.), sulfide-containing iron minerals (such as Schieleite, pyrite, etc.), and iron-containing spinels (such as zinc ferrite, copper ferrite, nickel ferrite, etc.); the inorganic ions include Cd. 2+ Co 2+ Cu 2+ Ni 2+ Pb 2+ Zn 2+ Inorganic cations and AsO3 2- AsO4 3- CrO4 2- Sb(OH)6 - SeO3 2- SeO4 2- Oxygen-containing anions.

[0015] Furthermore, in step S1, the relevant experimental conditions include parameters such as pH, temperature, ionic strength, initial concentration (adsorption rate prediction model), reaction time, and iron mineral concentration.

[0016] Further, in step S1, the properties of the iron mineral include parameters such as iron content, oxygen content, sulfur content, content of other metallic elements, specific surface area, zero charge point, adsorption amount (desorption rate prediction model), cell parameter a, cell parameter b, and cell parameter c; the properties of the inorganic ions include parameters such as relative atomic mass, element valence state, number of oxygen atoms, effective charge, element electronegativity, and ionic radius.

[0017] The data and parameters required to establish the dataset mentioned above are obtained from relevant literature, physical chemistry handbooks, databases or software (such as Materials Project, Crystallography Open Database, MDI Jade 6, etc.), and chemistry websites (such as https: / / ptable.com).

[0018] Furthermore, in step S1, the dataset for establishing the machine learning prediction model uses adsorption rate or desorption rate as the target variable and relevant experimental condition parameters, iron mineral property parameters and inorganic ion property parameters as input feature variables to construct adsorption dataset and desorption dataset respectively.

[0019] Furthermore, in step S2, the missing values ​​are filled using different filling methods according to the variable category. For example, the missing values ​​of variables related to iron mineral properties are filled using the average value of the corresponding variables in the same type of mineral, and the missing values ​​of variables related to experimental conditions are filled using the average value of the entire dataset.

[0020] Further, in step S2, the standardization method is as follows: The Z-score standardization method is used to transform the cleaned data into data with a mean of 0 and a standard deviation of 1. The calculation formula is:

[0021]

[0022] In the formula, For standardized values, These are the original sample values. This is the sample mean of this feature. This is the sample standard deviation of this feature.

[0023] Furthermore, in step S3, the correlation analysis is performed using the Spearman correlation coefficient, which is calculated as follows:

[0024]

[0025] In the formula, The Spearman correlation coefficient between the two features. For the first The rank difference of each sample This represents the number of samples.

[0026] Furthermore, in step S3, the importance analysis is based on the importance analysis method of random forest (RF). A preliminary adsorption rate and desorption rate prediction model is established using the random forest algorithm with default parameters, and then the feature importance values ​​of each model are calculated.

[0027] Further, in step S3, the method for removing features with strong relevance and low importance is as follows: when If a strong correlation is found between features, then importance analysis will be used to remove the feature importance values. Its characteristics.

[0028] Furthermore, in step S4, the different machine learning prediction models include K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), Extreme Random Tree (ET), Gradient Boosting Regression Tree (GBDT), and Extreme Gradient Boosting (XGB).

[0029] Furthermore, in step S4, the specific method for evaluating the predictive performance of each model is as follows:

[0030] The dataset processed in step S3 is divided into training and testing sets. Different machine learning prediction models are then used for model training to determine the optimal hyperparameter combination for each model. Using these optimal hyperparameter combinations, adsorption rate prediction models and desorption rate prediction models are constructed, respectively, and the coefficient of determination R0 is used. 2 The predictive performance of each model is evaluated using the mean absolute error (MAE) and the coefficient of determination (R²) on the test set. 2 The highest value, relatively low mean absolute error (MAE) value, and the coefficient of determination R between the training and test sets. 2 The model with the smaller value difference is selected as the optimal model; the coefficient of determination R... 2 The formulas for calculating the value and the mean absolute error (MAE) are as follows:

[0031]

[0032]

[0033] In the formula, For the sample size, This is the actual value. For predicted values, This is the average of the actual values.

[0034] More preferably, the method for determining the optimal hyperparameter combination for each model is as follows: first, the categories of hyperparameters to be optimized and their value ranges for each machine learning regression model are defined; then, the Optuna framework is used to automatically search for candidate hyperparameter combinations within a preset hyperparameter space, and the average R-value under five-fold cross-validation is used. 2 The value was used as an evaluation index for iterative optimization, and the average R was finally selected. 2 The hyperparameter combination with the highest value is taken as the optimal hyperparameter combination for the model.

[0035] Further, in step S5, the specific method for evaluating the binding stability is as follows: The data of the target iron mineral, inorganic ions, and set experimental conditions are input into the optimal adsorption and desorption rate model constructed in step S4 to obtain the corresponding adsorption and desorption rates. The adsorption-desorption retention amount RM of the iron mineral-inorganic ions is then calculated. A higher RM value indicates stronger binding stability. The formula for calculating the RM value is as follows:

[0036]

[0037]

[0038] In the formula, For desorption rate, This refers to the adsorption capacity. This represents the initial concentration of inorganic ions during the adsorption phase. For adsorption rate, This represents the concentration of iron minerals during the adsorption stage.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] (1) The present invention provides a machine learning-based method for evaluating the binding stability of iron minerals and inorganic ions. By collecting and organizing the experimental data of iron mineral-inorganic ion adsorption and desorption reported in the literature, a dataset containing information such as the properties of iron minerals, the properties of inorganic ions, and environmental conditions is constructed. The dataset is then preprocessed, feature-selected, and model-trained sequentially to establish an adsorption rate and desorption rate prediction model. Based on the established adsorption rate and desorption rate prediction model, the adsorption rate and desorption rate of the target iron mineral-inorganic ion system are predicted, and the adsorption-desorption retention RM is calculated to achieve a quantitative evaluation of the binding stability of the iron mineral-inorganic ion system.

[0041] (2) The original data required for this invention mainly comes from experimental results reported in existing literature. The data acquisition method is clear and existing research results can be fully utilized. The established adsorption and desorption rate prediction model can realize the rapid evaluation of the binding stability of different iron mineral-inorganic ion systems. It has the advantages of high prediction efficiency, wide applicability, and short evaluation cycle, which can save a lot of manpower and material resources. In addition, this invention avoids repeating complex adsorption-desorption experiments for each type of system to be evaluated, which can reduce the consumption of chemical reagents and the generation of experimental waste liquid, which meets the requirements of green and low-consumption development and has good application value. In addition, this invention is also expected to be applied to the analysis of environmental migration behavior of inorganic pollutants, environmental risk identification, pollution prevention and remediation scheme formulation, and the screening and performance evaluation of iron-based environmental remediation materials. Attached Figure Description

[0042] Figure 1 This is a flowchart of the steps of the machine learning-based method for evaluating the stability of iron mineral-inorganic ion binding in this invention.

[0043] Figure 2 The image shows the Spearman correlation coefficient heatmap in Example 1; where (a) is the Spearman correlation coefficient heatmap of the adsorption dataset and (b) is the Spearman correlation coefficient heatmap of the desorption dataset.

[0044] Figure 3 The graphs show the feature importance analysis results based on the random forest algorithm in Example 1; where (a) is the feature importance analysis result of the adsorption dataset and (b) is the feature importance analysis result of the desorption dataset.

[0045] Figure 4 The above are scatter plots of the actual and predicted values ​​of the optimal prediction model determined in Example 1; where (a) is a scatter plot of the actual and predicted values ​​of the optimal adsorption rate prediction model, and (b) is a scatter plot of the actual and predicted values ​​of the optimal desorption rate prediction model.

[0046] Figure 5 This is a heat map showing the binding stability of different iron mineral-inorganic ion systems in Example 1. Detailed Implementation

[0047] The present invention will be further described in detail below with reference to embodiments, but the implementation of the present invention is not limited thereto.

[0048] Example 1

[0049] A machine learning-based method for evaluating the stability of iron mineral-inorganic ion binding includes the following steps:

[0050] (1) Literature search was conducted in the Web of Science database, including three major search categories: interface processes, inorganic ions and iron minerals. The relevant search keywords are shown in Table 1.

[0051] Table 1. Search keywords for different search categories

[0052]

[0053] After searching database literature based on the aforementioned keywords, the adsorption and desorption rates of iron minerals and inorganic ions, as well as relevant experimental conditions, were collected from relevant literature. For data that could not be directly obtained from the literature text or tables, WebPlotDigitizer was used for data extraction. Iron mineral properties were obtained from databases or software such as the Materials Project, Crystallography Open Database, and MDI Jade 6, as well as relevant literature; inorganic ion properties were obtained from https: / / ptable.com, physical chemistry handbooks, and relevant literature. By integrating mineral properties, inorganic ion properties, experimental conditions, and corresponding adsorption or desorption rates, an iron mineral-inorganic ion adsorption and desorption dataset was constructed. The variables included in the adsorption and desorption datasets are shown in Table 2.

[0054] Table 2 shows the variables included in the adsorption and desorption datasets.

[0055]

[0056] (2) In the data collection and processing process of this embodiment, the original dataset is first checked for duplicate values. Some data may be duplicated due to repeated reports in different literature, or repeated recordings of the same experimental results in different charts and tables within the same literature. For example, pH adsorption edge data and adsorption isotherm data under corresponding pH conditions may contain the same experimental points. If such duplicate records are not identified and removed, certain iron mineral-inorganic ion systems or specific experimental conditions will be repeatedly counted in the dataset, thereby increasing their weight in model training and affecting the model's learning of the overall patterns. Therefore, in this embodiment, the original data is checked item by item according to information such as iron mineral type, inorganic ion type, and experimental conditions, and duplicate records are removed to improve the reliability of the dataset.

[0057] Furthermore, since the data used in this embodiment comes from different literatures, and the research objectives and test content of these literatures are not entirely consistent, some variables have missing values. To improve the completeness and usability of the dataset, missing values ​​were imputed before model construction. For variables related to iron mineral properties, since these variables mainly reflect the relatively stable physicochemical characteristics of the minerals themselves, similar iron minerals usually have certain similarities in these parameters. Therefore, the average value of the corresponding variable among similar iron minerals was used to impute the missing values. For variables related to environmental conditions, since these variables mainly depend on the specific experimental settings and do not have a fixed correspondence with a certain type of iron mineral or a certain type of inorganic ion, the average value of the corresponding variable in the overall dataset was used to impute the missing values.

[0058] Furthermore, since different input variables have different units of measurement and ranges of value, directly using them for model training may make some models more sensitive to variables with larger values. To mitigate the impact of differences in numerical range and units on model training, this embodiment uses the Z-score standardization method to process the input variables, converting each variable into standardized data with a mean of 0 and a standard deviation of 1. The formula for calculating Z-score standardization is as follows:

[0059]

[0060] In the formula, For standardized values, These are the original sample values. This is the sample mean of this feature. This is the sample standard deviation of this feature.

[0061] (3) Filter the input features in the dataset obtained in step (2) and use Spearman correlation analysis and random forest-based importance analysis to remove features with strong correlation and low importance.

[0062] First, correlation analysis is performed on the input features in the adsorption and desorption datasets, such as... Figure 2 As shown, darker colors represent stronger correlations. The absolute values ​​of the Spearman correlation coefficients for all feature pairs in the adsorption dataset are... The threshold was not reached, therefore the deletion of input features from the adsorption dataset was not considered in the correlation analysis. In the desorption dataset, iron content and oxygen content... Therefore, one of them needs to be deleted.

[0063] A random forest model is used to calculate and rank the importance of each input feature in the adsorption and desorption datasets, such as... Figure 3As shown in the figure, in the adsorption dataset, the importance values ​​of other metal element contents and oxygen atom counts are only 0.003 and 0.001, respectively, indicating low feature importance, and therefore they were deleted. In the desorption dataset, the importance values ​​of sulfur content and other metal element contents are both below 0.002, so these two features were deleted. In addition, due to the strong correlation between iron content and oxygen content, the relatively less important iron content was deleted to reduce feature redundancy.

[0064] The number of input variables for the final adsorption dataset was reduced from 21 to 19, and the number of input variables for the desorption dataset was reduced from 21 to 18.

[0065] (4) The data in the adsorption and desorption datasets were divided into training set and test set in an 8:2 ratio. Then, six regression algorithms, namely K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), Extreme Random Tree (ET), Gradient Boosting Decision Tree (GBDT) and Extreme Gradient Boosting (XGB), were used to train the model on the training set.

[0066] During model training, for each algorithm, the categories of hyperparameters to be optimized and their value ranges are first defined, as shown in Table 3. Then, the Optuna framework is used to automatically search and optimize hyperparameter combinations within a predefined hyperparameter space. Specifically, for each candidate hyperparameter combination, a five-fold cross-validation method is used to train and validate the model, and the average R-value of the candidate hyperparameter combination under five-fold cross-validation is calculated. 2 Value; the Optuna framework can determine the average R value based on the combination of candidate hyperparameters. 2 The value is used to evaluate its merits, with the average R value. 2 The highest value is taken as the optimization objective, and the search direction of subsequent hyperparameter combinations is continuously adjusted to determine the optimal hyperparameter combination of the algorithm. In this embodiment, the optimal hyperparameters of each algorithm in the adsorption rate and desorption rate prediction tasks are shown in Table 4.

[0067] Table 3. Hyperparameter search space of each algorithm

[0068]

[0069] Table 4 Optimal Hyperparameters for Each Algorithm

[0070]

[0071] By utilizing the optimal hyperparameters corresponding to each machine learning algorithm, adsorption rate prediction models and desorption rate prediction models are constructed respectively, and the coefficient of determination R is used. 2 The predictive performance of each algorithm model was evaluated using the mean absolute error (MAE), as shown in Table 5. (Coefficient of determination R...) 2The formulas for calculating the mean absolute error (MAE) are as follows:

[0072]

[0073]

[0074] In the formula, For the sample size, This is the actual value. For predicted values, This is the average of the actual values.

[0075] Table 5. Prediction performance of different algorithm models

[0076]

[0077] As shown in Table 5, the ET algorithm performs best in the adsorption rate prediction task, with R0 values ​​on both the training and test sets. 2 The R values ​​were 1.000 and 0.948, respectively, and the MAE values ​​were 0.002 and 4.772, respectively. In the desorption rate prediction task, the GBDT algorithm performed best, with R values ​​on both the training and test sets being [missing data]. 2 The values ​​were 0.977 and 0.917, respectively, and the MAEs were 2.855 and 5.317, respectively. Although in the desorption rate prediction task, the training set R of GBDT... 2 Slightly lower than XGB, MAE slightly higher than XGB, but its test set R 2 The highest value was obtained, and the difference between the training and test sets was smaller, indicating a relatively low degree of overfitting. Therefore, it performed best overall. Ultimately, the ET algorithm was chosen to construct the adsorption rate prediction model (ET model), and the GBDT algorithm was chosen to construct the desorption rate prediction model (GBDT model). The scatter plot of the actual and predicted values ​​is shown below. Figure 4 As shown.

[0078] (5) Based on the adsorption and desorption rate prediction model constructed in step (4), the adsorption and desorption rates of a typical iron mineral-inorganic ion system are predicted under the set experimental conditions. Then, the adsorption-desorption retention (RM) is calculated to evaluate the binding stability. The higher the retention, the stronger the binding stability. The calculation formula is:

[0079]

[0080]

[0081] In the formula, For desorption rate, This refers to the adsorption capacity. This represents the initial concentration of inorganic ions during the adsorption phase. For adsorption rate, This represents the concentration of iron minerals during the adsorption stage.

[0082] In this embodiment, to facilitate comparison and visualization of the differences in binding stability among different iron mineral-inorganic ion systems, the retention amount is normalized and converted into a dimensionless index with a value range of [0,1]. The specific calculation formula is as follows:

[0083]

[0084] In the formula, This represents the retention amount after normalization. For the first Retention amount of each iron mineral-inorganic ion system The maximum value of the retention amount in all systems. It represents the minimum retention value among all systems.

[0085] The binding stability of different iron mineral-inorganic ion systems was visualized based on the normalized retention amounts, and the results are as follows: Figure 5 As shown in the figure, the darker the color, the stronger the bonding stability.

[0086] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating the stability of a combination of iron minerals and inorganic ions based on machine learning, characterized by The method comprises the following steps: S1. Establishing a data set: collecting iron mineral-inorganic ion adsorption and desorption data and related experimental conditions, iron mineral properties and inorganic ion properties to establish a data set for machine learning prediction model; S2. Data preprocessing: cleaning the established data set, removing duplicate data and filling in missing values, and then standardizing the cleaned data; S3. Feature selection: performing correlation analysis and importance analysis on the variables in the data set, selecting input features, removing highly correlated and low importance features, and achieving feature dimension reduction; S4. Model construction: using different machine learning prediction models to predict the adsorption and desorption of iron minerals-inorganic ions based on the data set processed in step S3, and evaluating the prediction performance of each model to select the best prediction model for adsorption and desorption, respectively; S5. Combined stability evaluation: based on the best prediction model for adsorption and desorption constructed in step S4, the target iron mineral-inorganic ion system is predicted, and the adsorption-desorption retention RM is used to evaluate the combined stability of the target iron mineral-inorganic ion system.

2. The method for evaluating the stability of iron ore-inorganic ion combination based on machine learning according to claim 1, characterized in that: In step S1, the iron minerals include iron oxides or iron oxyhydroxides, sulfur-bearing iron minerals and iron-spinel; the inorganic ions include Cd 2+ , Co 2+ , Cu 2+ , Ni 2+ , Pb 2+ , Zn 2+ inorganic cations and AsO3 2- , AsO4 3- , CrO4 2- , Sb(OH)6 - , SeO3 2- , SeO4 2- oxygen-containing anions; the relevant experimental conditions include pH, temperature, ionic strength, initial concentration, reaction time and iron mineral concentration parameters; the iron mineral properties include iron content, oxygen content, sulfur content, other metal element content, specific surface area, point of zero charge, adsorption capacity, cell parameter a, cell parameter b and cell parameter c parameters; the inorganic ion properties include relative atomic mass, element valence, oxygen atom number, effective charge, element electronegativity and ion radius parameters. 3.The method of claim 1, wherein the method is characterized by: In step S1, the data set for machine learning prediction model takes adsorption or desorption as the target variable, and related experimental condition parameters, iron mineral property parameters and inorganic ion property parameters as input feature variables to construct adsorption data set and desorption data set, respectively.

4. The method for evaluating the stability of iron ore-inorganic ion combination based on machine learning according to claim 3, characterized in that: In step S2, the missing values are filled according to different filling methods based on variable categories, wherein the average value of the corresponding variables in the same mineral is used to fill the missing values of the iron mineral property related variables, and the average value of the overall data set is used to fill the experimental condition related variables; the standardization method is Z-score standardization method, which converts the cleaned data into data with mean value of 0 and standard deviation of 1, and the calculation formula is: wherein, is the standardized value, is the raw sample value, is the sample mean for this feature, is the sample standard deviation for this feature.

5. The method for evaluating the stability of iron minerals-inorganic ions combination based on machine learning according to claim 1, characterized in that: In step S3, the correlation analysis is performed using Spearman correlation coefficient, and the calculation formula is as follows: wherein is the Spearman's correlation coefficient between two features, is the rank difference of the i th sample, is the number of samples; The importance analysis is based on the importance analysis method of random forest, and the random forest algorithm with default parameters is used to establish the preliminary adsorption and desorption prediction model, and then the feature importance value of each model is calculated.

6. The method for evaluating the stability of iron ore-inorganic ion combination based on machine learning according to claim 5, characterized in that: The feature screening method for removing the features with strong correlation and low importance is as follows: when it is considered that the features have strong correlation, and the features with importance value are further deleted by combining the importance analysis method.

7. The method for evaluating the stability of iron ore-inorganic ion combination based on machine learning according to claim 1, characterized in that: In step S4, the different machine learning prediction models include K-nearest neighbor algorithm, support vector machine, random forest, extreme random tree, gradient boosting regression tree and extreme gradient boosting algorithm.

8. The method for evaluating the stability of iron ore-inorganic ion combination based on machine learning according to claim 7, characterized in that: The specific method for evaluating the prediction performance of each model is as follows: The dataset processed in step S3 is divided into training and testing sets. Different machine learning prediction models are then used for model training to determine the optimal hyperparameter combination for each model. Using these optimal hyperparameter combinations, adsorption rate prediction models and desorption rate prediction models are constructed, respectively, and the coefficient of determination R0 is used. 2 The predictive performance of each model is evaluated using the mean absolute error (MAE) and the coefficient of determination (R²) on the test set. 2 The highest value, relatively low mean absolute error (MAE) value, and the coefficient of determination R between the training and test sets. 2 The model with the smaller value difference is selected as the optimal model; the coefficient of determination R... 2 The formulas for calculating the value and the mean absolute error (MAE) are as follows: ， , where, is the number of samples, is the actual value, is the predicted value, is the mean of actual values.

9. The method for evaluating the stability of iron ore-inorganic ion combination based on machine learning according to claim 8, characterized in that: The method for determining the optimal hyperparameter combination of each model is: firstly, setting the hyperparameter categories to be optimized of each machine learning regression model and the value range thereof, then automatically searching the candidate hyperparameter combinations in the preset hyperparameter space by using an Optuna framework, and iteratively optimizing by using the average R 2 value under five-fold cross-validation as an evaluation index to finally select the hyperparameter combination with the highest average R 2 value as the optimal hyperparameter combination of the model.

10. The method for evaluating the stability of iron ore-inorganic ion combination based on machine learning according to claim 1, characterized in that: In step S5, the specific method for evaluating the combined stability is as follows: input the data of target iron mineral, inorganic ion and set experimental conditions into the best adsorption and desorption model constructed in step S4 to obtain the corresponding adsorption and desorption, and calculate the adsorption-desorption retention RM of iron mineral-inorganic ion, the higher the retention RM value, the stronger the combined stability; the calculation formula of RM value is as follows: ， wherein, is the desorption rate, is the adsorption capacity, is the initial concentration of inorganic ions in the adsorption stage, is the adsorption rate, is the concentration of iron minerals in the adsorption stage.

Images