Method for predicting micro-combination form of iron oxide and metal and application thereof

Abstract

The application provides a prediction method for iron oxide and metal micro combination form, S1, a data set for establishing a machine learning model; S2, based on the chemical connotation corresponding to the missing data in the data set, the missing data is filled, and a pretreated sample data set is obtained; S3, the pretreated sample data set is randomly divided into a training set and a test set; S4, an algorithm for machine learning is selected, hyperparameter optimization is performed on the training set, the optimal effect is selected, and the test set is evaluated, and an iron oxide and metal micro combination form prediction model is obtained; S5, the test sample data is substituted into the iron oxide and metal micro combination form prediction model, and an iron oxide and metal micro combination form prediction result is obtained. The problems that the prior art cannot provide coordination number and coordination bond length information on the micro level, the test operation of the advanced spectral instrument is complex, the cost is high, and the application universality is not possessed are solved.

Description

Technical Field

[0001] This invention relates to the field of pollution risk assessment technology, and in particular to a method for predicting the microscopic bonding morphology of iron oxides and metals and its application. Background Technology

[0002] Iron is the fourth most abundant element in the Earth's crust, and iron oxides (such as ferrihydrite, goethite, and hematite) are widely found in surface soil environments. At the same time, iron oxides play an important role in many industrial processes, such as hydrometallurgical zinc production, iron smelting, and wastewater treatment; therefore, they are also widely present in industrial solid waste.

[0003] Iron oxides mainly exist as crystalline or weakly crystalline nano- or micro-sized minerals or particles. The Fe atoms on the surface have unsaturated coordination sites, resulting in a large specific surface area and strong reactivity. They preferentially bind tightly to coexisting polymetallic compounds, thereby affecting the environmental stability, slow-release properties, and migration fate of other metals in industrial or geochemical processes.

[0004] Metals exhibit various bonding morphologies on iron oxides, such as electrostatic adsorption, inner sphere-side coordination, and inner sphere-angle coordination. The essence lies in the different coordination numbers and coordinate bond lengths of these microscopic bonding morphologies. For example, when antimony and iron coordinate along the same edge, their spacing is relatively short, approximately 3.1 Å, and they are microscopically more stable. When antimony and iron coordinate at an angle, their spacing is longer, approximately 3.55 Å, and the bonding is microscopically reversible. The degree of antimony bonding can be determined based on the coordination number between antimony and iron. Therefore, the coordination spacing and coordination number of metal microscopic bonding morphologies are fundamental to determining the macroscopic mobility, availability, and potential environmental risks of metals. Thus, elucidating the bonding morphologies of metals on iron oxides is crucial for predicting the fixation / release behavior of metals in the environment.

[0005] Currently, the main methods for determining the metal bonding morphology on iron oxides are experimental and spectroscopic methods. Experimental methods primarily rely on the adsorption behavior of metals to empirically determine the approximate bonding state, failing to provide microscopic information such as coordination number and bond length. Spectroscopic methods, such as extended X-ray absorption fine structure spectroscopy, can reflect the microscopic bonding state of metals more precisely; however, these advanced spectroscopic instruments have limited testing methods, are complex to operate, costly, and require specialized knowledge for spectral interpretation, limiting their widespread adoption and universality. Therefore, it is necessary to develop a method for predicting the microscopic bonding morphology of iron oxides with metals and its application, providing rapid and accurate predictions of the bonding morphology of iron oxides with metals, and offering convenient optimization guidance and refined technical support for the work processes of industries such as iron-containing and multi-metal systems, and environmental remediation. Summary of the Invention

[0006] The main objective of this invention is to provide a method for predicting the microscopic bonding morphology of iron oxides and metals, aiming to solve the problems of existing technologies being unable to provide microscopic information on coordination number and coordination bond length, and the complexity, high cost, and lack of universality of advanced spectroscopic instruments.

[0007] To achieve the above objectives, the present invention provides a method for predicting the microscopic bonding morphology of iron oxide and metal, characterized by comprising the following steps:

[0008] S1, The dataset for establishing the machine learning model; the dataset includes descriptor labels for the microscopic bonding morphology of iron oxides and metals obtained from publicly available literature, and the properties of the corresponding iron oxides, the properties of the metals, and the reaction conditions are used as input features.

[0009] S2, based on the chemical content corresponding to the missing data in the dataset, fill in the missing data to obtain a preprocessed sample dataset.

[0010] S3, randomly divide the preprocessed sample dataset into a training set and a test set.

[0011] S4. Select a machine learning algorithm, perform hyperparameter optimization on the training set, select the best performance and evaluate it on the test set to obtain the prediction model of the microscopic combination morphology of iron oxide and metal.

[0012] S5, Substitute the test sample data into the prediction model of the micro-bonding morphology of iron oxide and metal to obtain the prediction results of the micro-bonding morphology of iron oxide and metal.

[0013] Further, in step S1, the descriptor tags for the microscopic bonding morphology of the iron oxide and the metal include the bonding type, number of coordination configurations, coordination distance, and coordination number; the properties of the iron oxide include the type of iron oxide, iron content, iron-oxygen ratio, specific surface area, isoelectric point, cell parameters, and crystallinity; the properties of the metal include the atomic number, relative atomic mass, ionic radius, number of ionic oxygen atoms, valence state, number of outermost electrons, electronegativity, and surface charge; the reaction conditions include pH, background organic matter concentration, background ionic strength, temperature, time, and oxygen-containing conditions.

[0014] Furthermore, in step S2, when the missing data is discrete data, based on the chemical connotation corresponding to the missing data in the dataset, at least one of the following methods is used to fill in the missing data: standard substance property substitution method, similar reaction condition product property prediction method, and chemical equilibrium calculation method.

[0015] When the missing data is continuous data, based on the chemical content corresponding to the missing data in the dataset, at least one of the following methods is used to fill the missing data: DFT calculation method, mean imputation method, center metric method, K-nearest neighbor method, or EM algorithm.

[0016] Furthermore, between step S2 and step S3, the following steps are also included: sequentially performing encoding processing, feature continuity processing, standardization processing, and dimensionality reduction processing on the preprocessed sample data to obtain a dimensionality-reduced dataset.

[0017] Furthermore, the encoding process includes at least one of one-hot encoding, dummy encoding, and numerical encoding; the standardization process includes range standardization, linear proportional standardization, or Z-score standardization; and the dimensionality reduction process includes random forest feature importance, Lasso regression feature selection, correlation analysis, principal component analysis, autoencoder learning, or manifold learning.

[0018] Furthermore, the encoding process is a dummy encoding process; the dummy encoding process includes one-hot encoding, dummy encoding, or numerical encoding.

[0019] Further, in step S3, the step of randomly dividing the preprocessed sample dataset into a training set and a test set includes randomly dividing the dimensionality-reduced dataset into the training set and the test set.

[0020] Furthermore, in step S4, the machine learning algorithm includes one of gradient boosting tree, neural network, random forest, deep forest or support vector machine; the hyperparameter optimization specifically includes cross-validation using empirical methods, grid search methods or optimization algorithms.

[0021] The calculation formulas for the evaluation indicators are as follows:

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028] Where ACC is accuracy, SEN is sensitivity, SPC is specificity, R is correlation coefficient, MAE is mean absolute error, RMSE is root mean square error, TP is the number of true positive samples, TN is the number of true negative samples, P is the number of positive samples, and N is the number of negative samples. The actual value of the descriptor, The predicted value of the descriptor, The mean of the true values ​​in the dataset, These are the mean values ​​of the predicted values ​​in the dataset.

[0029] Further, in step S4, the machine learning algorithm includes the random forest or the deep forest; the hyperparameter optimization specifically includes performing the cross-validation using the optimization algorithm; wherein the optimization algorithm is the particle swarm optimization algorithm.

[0030] The present invention also provides an application of the prediction method for the microscopic combination morphology of iron oxide and metal as described in any of the above-mentioned methods in the industrial or environmental remediation of iron-containing and multi-metal systems.

[0031] The beneficial effects achieved by this invention are as follows:

[0032] This invention uses a predictive model to establish the microscopic bonding morphology of iron oxide and metal to predict the microscopic bonding morphology of iron oxide and metal, and is applicable to relevant processes in various industries and environmental remediation sectors.

[0033] 1. The forecasting is highly efficient, accurate, and flexible;

[0034] 2. It eliminates the need for complex experiments involving metal chemical reagents, avoids chemical pollution, conforms to the concept of green environmental protection, and is suitable for promotion;

[0035] 3. It overcomes the shortcomings of limited testing methods and complex, time-consuming operation of spectroscopic instruments. At the same time, it can quickly predict the microscopic combination morphology of iron oxide and metal under assumed conditions, thereby guiding and optimizing industrial processes. It is economical and practical. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0037] Figure 1 This is a flowchart of the implementation steps of Example 1;

[0038] Figure 2 This is a graph showing the importance analysis results for Example 1;

[0039] Figure 3 This is a graph showing the correlation analysis results for Example 1;

[0040] Figure 4 The following is a generalization ability evaluation chart of the prediction model in Example 1; where (a) is a comparison chart of the accuracy of the prediction of binding type and coordination configuration number in Example 1, and (b) is a comparison chart of R, RMSE and MAE of metal-iron coordination number and metal-iron spacing for different configurations in Example 1.

[0041] The realization of the objective, functional characteristics and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] It should be noted that, unless otherwise specified, the following embodiments and features can be combined with each other. It should also be understood that the terminology used in the embodiments of this invention is for describing specific implementations and not for limiting the scope of protection of this invention.

[0044] Unless otherwise defined, all technical and scientific terms used in this invention are consistent with the prior art known to those skilled in the art and the description of this invention. This invention can also be implemented using any prior art methods, devices, and materials similar to or equivalent to those described, used, or made of materials in the embodiments of this invention. It should be understood by those skilled in the art that, as an explanation of this application, without affecting the actual understanding of the technical solutions of this application, "max_depth" can represent the maximum depth of the model, "n_estimators" can represent the number of model learners, "n_trees" can represent the number of decision trees, "max_layers" can represent the maximum number of cascaded layers, "Accuracy" can represent the accuracy rate, "Sensitivity" can represent sensitivity, "Specificity" can represent specificity, "Adsorption" can represent adsorption reaction, "Oxic" can represent aerobic conditions, "Weak" can represent weak crystallinity, and "Strong" can represent strong crystallinity.

[0045] When numerical ranges are given in the examples, it should be understood that, unless otherwise stated in the invention, both endpoints of each range and any value between the two endpoints may be used. Test methods in the following examples that do not specify specific conditions are generally performed under conventional conditions or as recommended by the respective manufacturers. Unless otherwise specified, all materials or reagents required in the following examples are commercially available.

[0046] To address the shortcomings of existing technologies in providing microscopic information on coordination number and bond length, and the complex, costly, and limited applicability of advanced spectroscopic instruments, this invention provides a method for predicting the microscopic bonding morphology of iron oxides and metals, characterized by the following steps:

[0047] S1, the dataset for building the machine learning model; the dataset includes descriptor tags for the microscopic bonding morphology of iron oxides and metals obtained from publicly available literature, with the properties of the corresponding iron oxides, the properties of the metals, and the reaction conditions used as input features. Specifically, publicly available literature can be downloaded from platforms such as CNKI and Web of Science based on chemical principles, and descriptor tags for the microscopic bonding morphology of iron oxides and metals can be extracted from them.

[0048] S2, based on the chemical content corresponding to the missing data in the dataset, fill in the missing data to obtain a preprocessed sample dataset.

[0049] S3, randomly divide the preprocessed sample dataset into a training set and a test set.

[0050] S4. Select a machine learning algorithm, perform hyperparameter optimization on the training set, and evaluate the best performance on the test set to obtain a prediction model for the microscopic bonding morphology of iron oxide and metal.

[0051] S5, Substitute the test sample data into the prediction model of the micro-bonding morphology of iron oxide and metal to obtain the prediction results of the micro-bonding morphology of iron oxide and metal.

[0052] The above steps establish a predictive model for the microscopic bonding morphology of iron oxides and metals, which can be applied to relevant processes in various industries and environmental remediation sectors. The model offers high efficiency, accuracy, and flexibility; it eliminates the need for complex experiments involving metal chemical reagents, avoids chemical pollution, aligns with green environmental protection principles, and is suitable for widespread application; it also overcomes the shortcomings of spectroscopic instruments, such as limited testing methods, complex operation, and long processing times. Furthermore, it can quickly predict the microscopic bonding morphology of iron oxides and metals under assumed conditions, thereby guiding and optimizing industrial processes, demonstrating significant economic and practical advantages.

[0053] Furthermore, in step S1, the descriptor label for the microscopic bonding morphology of iron oxide and metal includes bonding type, number of coordination configurations, coordination spacing, and coordination number.

[0054] The properties of iron oxides include the type of iron oxide, iron content, iron-to-oxygen ratio, specific surface area, isoelectric point, cell parameters, and crystallinity. Specifically, the iron content and iron-to-oxygen ratio can be calculated using the standard chemical formula of the given mineral; the specific surface area of ​​iron oxides can be calculated by averaging the values ​​of different types of iron oxides; for cell parameters, the crystal structure data can be extracted from entries with similar reaction conditions (such as pH and temperature) by searching the PDF cards corresponding to the XRD results or by searching databases such as Springer Materials.

[0055] The properties of metals include atomic number, relative atomic mass, ionic radius, number of oxygen atoms in the ion, valence state, number of outermost electrons, electronegativity, and surface charge. Specifically, the surface charge can be calculated from pH and the pKa value of the metal ion; the radius of ions not specified in physical chemistry handbooks can be calculated using Gaussian quantum chemistry.

[0056] The reaction conditions include pH, background organic matter concentration, background ionic strength, temperature, time, and oxygen content. Specifically, the background ionic strength can be calculated using the characteristic ion concentrations of the extraction system; for temperature and oxygen content, average air temperature and atmospheric environmental data of the season preceding the submission date of the literature can be used.

[0057] Further, in step S2, when the missing data is discrete, it is filled using at least one of the following methods based on the chemical connotations of the missing data in the dataset: standard substance attribute substitution, similar reaction condition product attribute inference, and chemical equilibrium calculation. When the missing data is continuous, it is filled using at least one of the following methods based on the chemical connotations of the missing data in the dataset: DFT calculation, mean imputation, centrality metric, K-nearest neighbor method, or EM algorithm. Specifically, when filling missing data, for continuous data that is difficult to process, the read-across method, mean imputation, or K-nearest neighbor algorithm can be directly used; if multiple features are missing in a single data point, the data is deleted.

[0058] Furthermore, between steps S2 and S3, the following steps are also included: sequentially encoding the preprocessed sample data, performing feature continuity processing, standardization processing, and dimensionality reduction processing to obtain the dimensionality-reduced dataset.

[0059] Furthermore, the encoding process includes at least one of one-hot encoding, dummy encoding, and numerical encoding; the standardization process includes range standardization, linear proportional standardization, or Z-score standardization; and the dimensionality reduction process includes random forest feature importance, Lasso regression feature selection, correlation analysis, principal component analysis, autoencoder learning, or manifold learning.

[0060] Furthermore, the encoding process is a dummy encoding process; dummy encoding methods include one-hot encoding, dummy encoding, or numerical encoding.

[0061] Furthermore, in step S3, the preprocessed sample dataset is randomly divided into a training set and a test set, and the dimensionality-reduced dataset is randomly divided into a training set and a test set.

[0062] Specifically, dummy coding and feature continuation processing were performed on the discrete features in the complete dataset, increasing the number of input features to 31. Z-score standardization was then applied to the data. Feature importance and correlation analyses were performed on the input features. Features with high correlation were ranked by importance, retaining those with high importance and deleting those with low importance. Ultimately, the number of input variables was reduced to 15, thereby reducing redundant information and achieving higher modeling efficiency.

[0063] Furthermore, in step S4, the machine learning algorithm includes one of gradient boosting tree, neural network, random forest, deep forest or support vector machine; the hyperparameter optimization specifically includes cross-validation using empirical methods, grid search methods or optimization algorithms.

[0064] The evaluation indicators are calculated using the following formulas:

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071] Where ACC is accuracy, SEN is sensitivity, SPC is specificity, R is correlation coefficient, MAE is mean absolute error, RMSE is root mean square error, TP is the number of true positive samples, TN is the number of true negative samples, P is the number of positive samples, and N is the number of negative samples. For the actual value of the descriptor, The predicted value of the descriptor. The mean of the true values ​​in the dataset, These represent the mean of the predicted values ​​in the dataset. Specifically, to avoid randomness and overfitting in the results, a 5-fold cross-validation model for predicting the microscopic bonding morphology of iron oxides and metals can be used.

[0072] Further, in step S4, the machine learning algorithm includes random forest or deep forest; hyperparameter optimization specifically includes using an optimization algorithm and performing cross-validation; wherein the optimization algorithm is particle swarm optimization. Specifically, particle swarm optimization can be used to optimize hyperparameters on the training set.

[0073] This invention also provides an application of the prediction method for the microscopic bonding morphology of iron oxides and metals using any of the above-described methods in the industrial or environmental remediation of iron-containing and multi-metal systems. Through the above steps, the microscopic bonding morphology of iron oxides and metals in the industrial or environmental remediation of iron-containing and multi-metal systems can be predicted efficiently, accurately, and flexibly.

[0074] To further illustrate the present invention, the following examples are provided:

[0075] Example 1

[0076] It should be noted that the specific procedures for each implementation step are as follows: Figure 1 As shown.

[0077] 1. Based on the principle of coordination equilibrium, relevant literature was downloaded from CNKI and Web of Science, and descriptor tags for a series of binding forms, such as binding type, number of coordination configurations, coordination interval, and coordination number, were extracted.

[0078] Specifically, the binding type of the metal is obtained from spectral results such as extended X-ray absorption fine structure spectroscopy or electron pair distribution function, determining whether a metal-iron coordination path exists. Then, for those with metal-iron coordination, the number of configurations for different metal-iron coordinations (usually single or double configurations) is collected. Finally, the coordination distance and coordination number are obtained for each coordination configuration. The attributes of each descriptor tag are shown in Table 1.

[0079] Table 1. Attributes of descriptor tags for metal microstructure bonding.

[0080]

[0081] Relevant literature was downloaded from CNKI and Web of Science to obtain the properties and reaction conditions of iron oxides that guide specific metal bonding morphology descriptors. These included iron oxide types, iron content, iron-to-oxygen ratio, specific surface area, isoelectric point, cell parameters, crystallinity, pH, background organic matter concentration, background ionic strength, temperature, time, and oxygen-containing conditions. The relative atomic mass, valence state, surface charge, ionic radius, electronegativity, and number of ionic oxygen atoms of the metals were obtained through literature, physicochemical databases, and chemical calculations. The above input feature data was combined with descriptor label data to form the dataset for the machine learning model. Detailed attributes of each input feature are shown in Table 2.

[0082] Table 2. Influencing Factors and Characteristics of Metal Microstructure Bonding

[0083]

[0084] 2. The missing data in the dataset is filled in based on the chemical connotations corresponding to the missing data. Specifically, iron content and iron-oxygen ratio are calculated using the standard chemical formulas of the given minerals; the specific surface area of ​​iron oxides is calculated as an average value based on the different types of iron oxides; for unit cell parameters, the PDF cards corresponding to the XRD results are searched, or databases such as Springer Materials are searched to extract crystal structure data from entries with similar formation conditions (such as pH, temperature, etc.); the surface charge of metals is calculated based on pH and the pKa value of metal ions; the radius of ions not specified in the physical chemistry handbook is calculated using Gaussian quantum chemistry; the ionic strength is calculated using the characteristic concentration of the extracted ions; for temperature and oxygen conditions, in this example, the average temperature and atmospheric environmental data of the season preceding the submission time of the literature's location are used. For other continuous data, if it is difficult to use similar methods as described above, a read-across method is used for filling; if multiple features are missing in a single data point, that data is deleted; thus, a preprocessed sample dataset is obtained.

[0085] 3. For the discrete features in the preprocessed sample dataset, dummy coding and feature continuation were performed sequentially, increasing the number of input features to 31. Z-score standardization was then applied to the data. Feature importance and correlation analyses were performed on the input features. Features with high correlation were ranked by importance, retaining those with high importance and deleting those with low importance. Finally, the dimensionality of the input variables was reduced to 15, resulting in the sample dataset. The importance analysis results are shown in the figure below. Figure 2 As shown in the figure; the correlation analysis results are shown in the figure below. Figure 3 As shown.

[0086] The sample dataset was randomly divided into a training set and a test set, with the training set accounting for 80% of the total sample dataset and the test set accounting for 20%.

[0087] 4. Random forest was selected as the machine learning algorithm. Evaluation metrics for discrete descriptor prediction included accuracy (ACC), sensitivity (SEN), and specificity (SPC); evaluation metrics for continuous descriptor prediction included correlation coefficient (R), mean absolute error (MAE), and root mean square error (RMSE).

[0088] Particle swarm optimization (PSO) was used to optimize hyperparameters on the training set. The population size was set to 100, and the PSO iteration count was set to 200. To avoid randomness and overfitting, 5-fold cross-validation was employed. The optimal hyperparameters for the iron oxide-metal microstructure prediction model are shown in Table 3.

[0089] Table 3. Optimal hyperparameter combination for the prediction model of microscopic bonding morphology between iron oxides and metals.

[0090]

[0091] 5. Finally, the predicted results of the microscopic bonding morphology of iron oxide and metal in the test set were obtained.

[0092] The generalization ability of the prediction model for the microscopic bonding morphology of iron oxides and metals was evaluated. The evaluation results of the deep forest prediction model are as follows: Figure 4 As shown. According to Figure 4 Observations show that the accuracy rates for predicting the bonding type and coordination configuration number of metals are 98.8% and 91.7%, respectively. Figure 4 As shown in (a), the predicted R-values ​​for the metal-iron coordination number and spacing of the first configuration and the metal-iron coordination number and spacing of the second configuration are 0.838, 0.903, 0.788, and 0.868, respectively. Figure 4 As shown in (b).

[0093] The above results show that the prediction model for the micro-bonding morphology of iron oxide and metal provided by this invention has excellent generalization prediction ability when predicting the micro-bonding morphology of iron oxide and metal, indicating that the prediction method for the micro-bonding morphology of iron oxide and metal is effective and feasible.

[0094] In summary, the above-described technical solutions of the present invention are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. All equivalent structural transformations made using the contents of the present invention's specification and drawings under the technical concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method of predicting an iron oxide and metal micro-association form, characterized by, The method comprises the steps of: S1, establishing a data set of a machine learning model; the data set comprises obtaining a descriptor label of a microcosmic combination form of iron oxide and metal from a public literature, and taking properties of the corresponding iron oxide, properties of the metal, and reaction conditions as input features; The descriptor label of the microcosmic combination form of iron oxide and metal comprises a combination type, a coordination configuration number, a coordination spacing, and a coordination number; the properties of the iron oxide comprise an iron oxide type, an iron content, an iron oxygen ratio, a specific surface area, an isoelectric point, a cell parameter, and a crystallinity; the properties of the metal comprise an atomic number, a relative atomic mass, an ionic radius, an ionic oxygen atom number, a valence state, an outermost electron number, electronegativity, and a surface charge; and the reaction conditions comprise pH, background organic matter concentration, background ion strength, temperature, time, and oxygen-containing conditions; S2, filling missing data in the data set based on chemical connotations corresponding to the missing data, to obtain a preprocessed sample data set; S3, randomly dividing the preprocessed sample data set into a training set and a test set; S4, selecting a machine learning algorithm, performing hyperparameter optimization on the training set, selecting an optimal effect, and evaluating the optimal effect on the test set, to obtain a microcosmic combination form prediction model of iron oxide and metal; S5, substituting test sample data into the microcosmic combination form prediction model of iron oxide and metal, to obtain a microcosmic combination form prediction result of iron oxide and metal.

2. The method of predicting iron oxide and metal micro-association forms according to claim 1, characterized in that, In step S2, when the missing data is discrete data, at least one of a standard substance attribute substitution method, a similar reaction condition product attribute inference method, and a chemical equilibrium calculation method is used to fill the missing data based on the chemical connotations corresponding to the missing data in the data set; when the missing data is continuous data, at least one of a DFT calculation method, a mean filling method, a central metric value method, a K-nearest neighbor method, or an EM algorithm is used to fill the missing data based on the chemical connotations corresponding to the missing data in the data set.

3. The method of predicting iron oxide and metal micro-associations according to claim 1, characterized in that, Between the step S2 and the step S3, the preprocessed sample data is sequentially subjected to encoding processing, feature continuous processing, standardization processing, and dimension reduction processing, to obtain a dimension-reduced data set.

4. The method of predicting iron oxide and metal micro-association forms according to claim 3, characterized in that, The encoding processing comprises at least one of one-hot encoding, dummy encoding, or numerical encoding; The standardization processing comprises a range standardization method, a linear proportion standardization method, or a Z-score standardization method; The dimension reduction processing comprises a random forest feature importance, a Lasso regression feature selection, a correlation analysis method, a principal component analysis method, a self-encoder learning, or a flow learning.

5. The method of predicting iron oxide and metal micro-association forms according to claim 4, characterized in that, The encoding processing is dummy encoding processing; the dummy encoding processing mode comprises one-hot encoding, dummy encoding, or numerical encoding.

6. The method of predicting iron oxide and metal micro-associations of claim 1, wherein, In step S3, the preprocessed sample data set is randomly divided into the training set and the test set, which comprises randomly dividing the dimension-reduced data set into the training set and the test set.

7. The method of predicting iron oxide and metal micro-associations according to claim 1, characterized in that, In step S4, the machine learning algorithm comprises one of a gradient boosting tree, a neural network, a random forest, a deep forest, or a support vector machine; The hyperparameter optimization specifically includes adopting an experience method, a grid search method or a particle swarm algorithm to perform cross-validation. The index calculation formula of the evaluation is as follows: wherein ACC is accuracy, SEN is sensitivity, SPC is specificity, R is correlation coefficient, MAE is mean absolute error, RMSE is root mean square error, TP is the number of true positive samples, TN is the number of true negative samples, P is the number of positive samples, and N is the number of negative samples, is a true value of the descriptor, is a predicted value of the descriptor, is a mean of true values in the data set, is a mean of predicted values in the data set, respectively.

8. The method of predicting iron oxide to metal micro combinations of claim 1, wherein, In step S4, the algorithm of the machine learning includes a random forest or a deep forest.

9. Use of the method for predicting the micro-association form of iron oxides and metals according to any one of claims 1 to 8 in the industrial or environmental management of iron and polymetallic systems.

Images