A fenton-like reaction bimetallic catalyst screening method, device and storage medium based on machine learning and dft theory calculation guidance

Abstract

This invention relates to the field of advanced oxidation technology mediated by bimetallic catalysts, specifically to a method, apparatus, and storage medium for screening Fenton-like bimetallic catalysts based on machine learning and DFT theoretical calculations. This invention aims to address the inefficiencies and blind spots inherent in traditional experimental trial-and-error methods for the synthesis and optimization of bimetallic single-atom catalysts. It provides a Fenton-like bimetallic catalyst screening method based on machine learning and DFT theoretical calculations to achieve efficient degradation of pollutants in water. This method can accurately identify key design parameters, verify the catalytic reaction mechanism at the electronic scale, and guide the controllable synthesis of catalysts, thereby significantly improving material development efficiency and demonstrating broad application prospects.

Description

Technical Field

[0001] This invention relates to the field of advanced oxidation technology mediated by bimetallic catalysts, and specifically to a method, apparatus, and storage medium for screening Fenton-like bimetallic catalysts based on machine learning and DFT theoretical calculations. Background Technology

[0002] Advanced oxidation technologies based on sulfate radicals are effective means of treating recalcitrant organic pollutants in water bodies. Among them, the Fenton-like reaction, which activates persulfate (PMS) or perdisulfate (PDS) with transition metal catalysts to generate highly oxidizing radicals, has become a research hotspot due to its rapid reaction and wide pH adaptability. Compared with monometallic catalysts, bimetallic catalysts often significantly improve the activation efficiency and stability of oxidants through synergistic effects between metals (such as electron transfer and redox cycles). Therefore, developing efficient bimetallic catalysts is key to enhancing the engineering application potential of Fenton-like technologies.

[0003] However, the design of bimetallic catalysts is a typical high-dimensional multi-objective optimization problem, and their performance is complexly influenced by multiple factors. These factors include: 1) intrinsic electronic structure factors: such as the types and ratios of the two active metal elements (directly affecting the average number of d electrons) and the electronic properties of the support material (number of outermost electrons); 2) material preparation process parameters: such as precursor composition, heat treatment temperature and time, etc., which determine the catalyst's crystal phase, size, dispersion, and metal-support interaction; 3) reaction operating conditions: such as reaction temperature, pH value, pollutant and oxidant concentrations, catalyst dosage, etc. Traditional trial-and-error methods or R&D models relying on limited experience require repeated experiments in a vast multi-dimensional parameter space, resulting in long R&D cycles, huge resource consumption, and difficulty in revealing deep structure-activity relationships, severely restricting the efficiency of new material development.

[0004] In recent years, the integration of computational materials science and artificial intelligence has brought new ideas to materials research and development. On the one hand, machine learning (ML) can extract complex nonlinear relationships from massive experimental or computational data, establish "structure / process-performance" predictive models, and achieve rapid performance prediction. On the other hand, density functional theory (DFT) calculations can accurately calculate the electronic structure, adsorption energy, reaction pathways, etc. of materials at the atomic and electronic levels, providing microscopic insights into understanding catalytic mechanisms. By combining machine learning with theoretical calculations and experimental verification, universal design principles can be extracted from complex data to guide the development and optimization of high-performance bimetallic catalysts.

[0005] Currently, existing studies have employed machine learning methods to predict catalyst performance or relied solely on density functional theory (DFT) calculations for material screening. However, research on the deep integration of these two methods in the design of Fenton-like bimetallic catalysts remains relatively limited.

[0006] Therefore, developing a screening method for Fenton-like bimetallic catalysts based on machine learning and DFT calculations is of great significance for promoting the development of advanced oxidation technologies mediated by bimetallic catalysts. Summary of the Invention

[0007] This invention aims to address the inefficiencies and blind spots inherent in traditional experimental trial-and-error methods for the synthesis and optimization of bimetallic single-atom catalysts. It provides a Fenton-like reaction bimetallic catalyst screening method guided by machine learning and DFT theoretical calculations to achieve efficient degradation of pollutants in water. This method can accurately identify key design parameters, verify the catalytic reaction mechanism at the electronic scale, and guide the controllable synthesis of catalysts, thereby significantly improving material development efficiency and demonstrating broad application prospects.

[0008] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a screening method for Fenton-like reaction bimetallic catalysts based on machine learning and DFT theoretical calculations, comprising: S1. Obtain data on the degradation of pollutants in water by bimetallic catalysts activating oxidants in Fenton-like reactions; The data information includes: the average number of d electrons in the transition metal of the bimetallic catalyst, the number of outermost electrons in the support of the bimetallic catalyst, the catalyst preparation conditions, and the pollutant degradation reaction conditions. S2. Using the data from S1 as input and the degradation rate constant as the target feature, construct a pollutant degradation rate constant prediction model based on the XGBoost algorithm; S3. Set the expected degradation rate constant k value to obtain the average d electron number of the transition metal in the bimetallic catalyst, the number of outermost electrons of the support, and the catalyst preparation conditions; S4. Construct the molecular structure of the bimetallic catalyst based on the average number of d electrons of the transition metal and the number of outermost electrons of the support, and perform DFT theoretical calculations to obtain the adsorption energy of the bimetallic catalyst with different metal element compositions for the oxidant. S5. Select the metal element composition of the bimetallic catalyst based on the adsorption energy of S4, and determine the preparation conditions of the bimetallic catalyst based on the catalyst preparation conditions obtained in S3.

[0009] Catalyst properties, preparation conditions, and reaction conditions are three different dimensions describing advanced oxidation processes mediated by bimetallic catalysts. Optimization of a single dimension cannot fully reflect the actual degradation performance under the coupled effects of multiple factors, making it difficult to effectively guide the design and optimization of highly efficient catalysts. Machine learning techniques can integrate these multidimensional characteristics, revealing the complex nonlinear relationship between them and degradation performance. Simultaneously, combined with DFT theoretical calculations, the ability of different materials to activate oxidants can be compared at the electronic level. Based on these results, experimental schemes for the preparation and degradation of bimetallic catalytic materials can be formulated, significantly improving the efficiency of material development in the field of antibiotic degradation.

[0010] Specifically, the XGBoost algorithm was used to construct a model using 559 sets of degradation data collected from literature as both training and testing sets. During training, the relationships between catalyst characteristics, preparation conditions, reaction conditions, and degradation rate constants were established to obtain a global optimization model and identify key descriptors. Adsorption energy and density of states were calculated using DFT theory to verify the regulatory role of descriptors in the oxidant activation mechanism. The characteristic parameters identified by the model can be used to specifically adjust material synthesis and optimization schemes without the need for trial-and-error experiments to select the optimal conditions. This significantly reduces experimental costs and time consumption, and can effectively guide catalyst design.

[0011] Therefore, the method established in this invention can guide the screening of Fenton-like bimetallic catalysts based on machine learning and DFT theoretical calculations, thus obtaining a more efficient and universal material development path.

[0012] Preferably, in S1, the catalyst preparation conditions include heat treatment temperature and heat treatment time.

[0013] And / or, the pollutant degradation reaction conditions include reaction temperature, initial pH value, initial pollutant concentration, catalyst dosage, and oxidant dosage.

[0014] And / or, the contaminant is an antibiotic.

[0015] And / or, the oxidant is one of persulfate, perdisulfate or hydrogen peroxide.

[0016] Preferably, in S2, the data information is input after being standardized to a uniform scale space by Z-score.

[0017] Preferably, the Z-score standardization process uses the StandardScaler tool provided by the scikit-learn library.

[0018] Literature data from various sources covered degradation performance under diverse catalytic systems and operating parameters. The obtained data were randomly divided into training and test sets in an 8:2 ratio. The StandardScaler tool provided by the scikit-learn library was used to map the original feature variables to a uniform scale space with a mean of 0 and a standard deviation of 1 through Z-score standardization, thus eliminating dimensional differences and ensuring the numerical stability, statistical consistency, and interpretability of each feature weight during model training.

[0019] Preferably, in S2, the pollutant degradation rate constant prediction model uses grid search and five-fold cross-validation to optimize hyperparameters; the optimized parameters include: the number of decision trees is set to 50~200, the maximum depth is set to 3~9, the learning rate is set to 0.01~0.2, the subsampling ratio is set to 0.6~1.0, the column sampling ratio is set to 0.6~1.0, the L1 regularization parameter is set to 0~0.5, and the L2 regularization parameter is set to 1~2.

[0020] The model was evaluated on the test set by comparing the predicted values ​​with the actual values ​​of five models: KNN, RF, XGBoost, LightGBM, and Transformer. The XGBoost model performed better in R... 2 It performed best in terms of RMSE and MAE, verifying its applicability in this invention.

[0021] Preferably, in S3, after obtaining the target feature parameters, the prediction model is further analyzed and interpreted using SHAP values ​​to quantify the contribution of each input feature in S2 to the degradation rate constant k; and the nonlinear influence of key input features on the degradation rate constant k during catalyst design and degradation experiments is revealed using PDP plot analysis.

[0022] Preferably, in S4, the specific steps of the DFT theoretical calculation are as follows: S4.1. Using Gaussian software, under the theoretical level of B3LYP / def2SVP, the geometric structure and vibration frequency of the catalytic material before and after adsorption of oxidant were optimized and calculated. GD3BJ dispersion correction was introduced. Cartesian coordinates and the keywords Calcfc and NoEigenTest were used in the optimization process to ensure convergence stability. S4.2. Based on the stable structure optimized in S4.1, the single-point energy is calculated using the B3LYP / 6-311++G(d,p) method combined with GD3BJ dispersion correction.

[0023] Preferably, in S4.1, the keyword combination of the input file in the optimization calculation and vibration frequency calculation is opt=(calcfc,noeigentest,cartesian) freq b3lyp def2svp em=gd3bj.

[0024] The choice of basis sets and functionals in theoretical calculations determines the "physical reality" of the potential energy surface; selecting an appropriate computational level can improve computational efficiency and accuracy. The HF / 3-21G level is generally considered "inaccurate" when quantitative results are required or when describing weak interactions and electron correlation effects; under the B3LYP functional condition, def2SVP is chosen instead of 6-311G. This is because def2SVP automatically invokes pseudopotentials for heavy elements such as transition metals, and its convergence behavior is more stable in actual calculations, avoiding the 6-311G problem. The non-convergence problem often occurs due to the inappropriate description of heavy atoms by the all-electron basis set. At the same time, this combination has been extensively verified in catalysis and materials systems, and the results are reliable.

[0025] Preferably, in S5, based on the adsorption energy of S4, the Multiwfn software is used to perform d-orbital projected density of states analysis according to the output file of the Gaussian software to clarify the adsorption behavior and activation mechanism between bimetallic d orbitals and between metal d orbitals and oxidant, thereby screening out the metal element components of bimetallic catalysts with stronger oxidant activation ability.

[0026] Preferably, the metal element composition includes the metal element composition and proportion.

[0027] This invention also provides a device for predicting the composition of Fenton-like bimetallic catalyst materials based on machine learning and DFT theoretical calculations, comprising: The data acquisition module is used to acquire data on the degradation of pollutants in water by bimetallic catalysts activating oxidants in Fenton-like reactions. The catalyst design module is used to output the average number of d electrons of the transition metal, the number of outermost electrons of the support, and the catalyst preparation conditions based on the set expected degradation rate constant k value using the trained XGBoost prediction model. The DFT calculation and screening module is used to calculate the adsorption energy of bimetallic catalysts with different metal element compositions, and to determine the metal element composition of the bimetallic catalyst based on the calculation results.

[0028] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a method for predicting the composition of a Fenton-like bimetallic catalyst material based on machine learning and DFT theoretical calculations.

[0029] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for predicting the composition of a Fenton-like bimetallic catalyst material guided by machine learning and DFT theoretical calculations.

[0030] Therefore, the present invention has the following beneficial effects: (1) This invention utilizes machine learning to integrate multi-dimensional data such as catalyst characteristics, preparation conditions and reaction conditions, and identifies key descriptors through SHAP and partial dependency graphs to clarify the optimal design range in catalyst synthesis, providing accurate synthesis basis for the selection of metal types and processing technology of bimetallic single-atom catalysts, and avoiding the blind screening of traditional trial and error methods.

[0031] (2) This invention uses DFT theory to calculate electronic structure indicators such as adsorption energy and density of states, and verifies the regulatory role of key descriptors identified by machine learning on the oxidant activation mechanism from an electronic perspective. It combines macroscopic performance prediction with microscopic mechanism understanding, providing theoretical support for catalyst synthesis and optimization. Based on the theoretical calculation results, the catalyst metal combination is selected to ensure that the prepared material has a high efficiency in oxidant activation.

[0032] (3) Based on the optimal design range identified by machine learning and the electronic structure verified by DFT, this invention provides precise guidance for the preparation of bimetallic single-atom catalysts by solution-gel-calcination method. It also allows for targeted adjustment of metal combinations, ratios, and synthesis variables such as heat treatment temperature and time, thereby significantly improving material development efficiency and reducing experimental costs.

[0033] (4) The present invention verified the efficient degradation ability of the catalyst synthesized under the guidance pathway to antibiotics in water through degradation experiments. It provides a universal and reproducible technical solution for the directional design and controllable synthesis of efficient bimetallic single-atom catalysts, which has important application value and market potential. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of a screening method for Fenton-like bimetallic catalysts based on machine learning and DFT theoretical calculations.

[0035] Figure 2 The training results of the XGBoost model are shown in (a) comparison between predicted and actual values, and (b) SHAP analysis.

[0036] Figure 3 The following are partial feature PDP plots: (a) PMS, (b) catalyst dosage, (c) two-dimensional PDP plot of PMS and catalyst dosage, (d) average d electron number of catalyst transition metal, (e) heat treatment temperature, and (f) two-dimensional PDP plot of average d electron number of catalyst transition metal and heat treatment temperature.

[0037] Figure 4 The images show the optimized structure of the catalyst after PMS adsorption: (a) Co-Cu, (b) Cu-Mn.

[0038] Figure 5 This is a projection density of states of the 3d orbitals of the catalyst transition metal and the 2p orbitals of the O atom on PMS.

[0039] Figure 6 The activation of PMS as a catalyst material to degrade antibiotics. Detailed Implementation

[0040] The present invention will be further described below with reference to specific embodiments. Those skilled in the art will be able to implement the present invention based on these descriptions. Furthermore, the embodiments of the present invention described below are generally only some, not all, of the embodiments of the present invention. Therefore, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0041] Example 1 A screening method for Fenton-like reaction bimetallic catalysts based on machine learning and DFT theoretical calculations includes: S1. Obtaining Data Information S1.1 Search for relevant published papers that match the research topic from literature search websites such as Web of Science and Google Scholar using keywords such as "Bimetal", "Peroxymonosulfate (PMS)" and "Antibiotic degradation".

[0042] S1.2 Catalyst Feature Extraction: Extract relevant data on the average number of d electrons (n_d) of the catalyst transition metal and the number of outermost electrons (n_outer) of the catalyst support from the screened papers.

[0043] S1.3 Extraction of catalyst preparation conditions: Extract relevant data on heat treatment temperature (HT_temp) and heat treatment time (HT_time) during catalyst preparation from the screened papers.

[0044] S1.4 Extraction of pollutant degradation reaction conditions: Data on antibiotic degradation reaction temperature (R_temp), initial pH value (pH), initial antibiotic concentration (Antibiotic), catalyst dosage (Catalyst), and PMS dosage (PMS) were extracted from the screened papers.

[0045] S1.5 Target Feature Extraction: Extract relevant data on the antibiotic degradation rate constant k from the screened papers.

[0046] S1.6 Entries with missing values ​​were deleted. Finally, a complete benchmark dataset consisting of 559 rows and 10 columns was obtained. Statistical analysis was performed on the collected benchmark dataset (Table 1) to ensure the accuracy and validity of subsequent analyses. First, descriptive statistical analysis was performed as a foundational step, outlining the distribution characteristics of each variable within the dataset; then, a Pearson correlation heatmap was used to quantify and visualize the correlation strength between variables, retaining weak correlations for model construction.

[0047] Table 1 Data Distribution

[0048] S2. Machine Learning Model Building and Training S2.1 Data Preprocessing and Model Training: The obtained data is randomly divided into training and test sets in an 8:2 ratio. The StandardScaler tool provided by the scikit-learn library is used to map the original feature variables to a uniform scale space with a mean of 0 and a standard deviation of 1 through Z-score standardization, so as to eliminate the difference in units and thus ensure the numerical stability, statistical consistency and interpretability of each feature weight in the model training.

[0049] S2.2 Parameter Super-Optimization: Hyperparameters were optimized using grid search combined with five-fold cross-validation. The optimal hyperparameter model for XGBoost was: number of decision trees (n_estimators) = 200, maximum depth (max_depth) = 7, learning rate (learning_rate) = 0.2, subsample ratio (subsample) = 0.8, column sample ratio (colsample_bytree) = 1.0, L1 regularization parameter (reg_alpha) = 0.1, and L2 regularization parameter (reg_lambda) = 2.

[0050] S2.3 Model Evaluation: The model is evaluated on the test set. The results of comparing the predicted and actual values ​​of the XGBoost model are as follows: Figure 2 As shown in (a). Model performance evaluation metrics include the coefficient of determination (R²). 2The root mean square error (RMSE) and mean absolute error (MAE) are calculated using the following formulas:

[0051]

[0052]

[0053] in, The actual degradation rate constant k (true value) was measured in the experiment of the i-th catalyst group. Let k be the predicted degradation rate constant (predicted value) for the i-th catalyst group. Let k be the mean of all true degradation rate constants, and N be the number of samples.

[0054] Depend on Figure 2 (a) and Table 3 show that the trained XGBoost model R 2 =0.725, RMSE=0.080, MAE=0.047, indicating that the model has excellent prediction accuracy and generalization ability.

[0055] S3. Catalyst Design S3.1 The expected degradation rate constant k is input into the trained XGBoost model, and the contribution of each input feature to the prediction result is analyzed using the SHAP method. The results are as follows: Figure 2 As shown in (b). The results show that the PMS concentration has the greatest impact on the k value, followed by the average number of d electrons of the transition metal in the bimetallic catalyst, the heat treatment temperature of catalyst preparation, the amount of catalyst added, the heat treatment time of catalyst preparation, the reaction temperature, the initial pH value of the reaction, the initial concentration of antibiotic, and the number of outermost electrons of the bimetallic catalyst support.

[0056] S3.2 uses PDP plots and two-dimensional PDP plots to analyze the nonlinear relationship between the four features with the highest contribution values ​​obtained in S3.2 and the degradation rate constant k from two perspectives: pollutant degradation reaction conditions and catalyst preparation conditions. The results are as follows: Figure 3 As shown. The data was standardized before model training, so the actual value X corresponding to each feature variable is calculated according to the following formula.

[0057]

[0058] Where std is the standard deviation of the original data, Z is the standardized value, and mean is the mean of the original data.

[0059] Based on the PDP analysis results, it was found that the k value increased sharply when the PMS concentration was 1.5 mM, and then the k value increased with the increase of PMS dosage, stabilizing at 3.3 mM. In the catalyst dosage range of 0.01~0.5 g / L, the k value increased with the increase of dosage, and then stabilized. The average d electron number of the catalyst transition metal in the range of 5.5~8.7 corresponds to the peak k value. With the increase of heat treatment temperature, the k value continued to increase.

[0060] Based on the PDP analysis results, an optimized scheme for catalytic material preparation and reaction conditions was formulated to maximize the degradation rate constant k value: In terms of material preparation, the heat treatment temperature was set to 800℃, and the average d electron number of the catalyst transition metal was controlled within the peak range of 5.5~8.7; in terms of reaction conditions, the PMS dosage was set to 1.5mM (this concentration is the turning point where the k value rises sharply), and the catalyst dosage was set to 0.25g / L, which served as the integrated initial optimized conditions for achieving efficient antibiotic degradation.

[0061] S4. DFT Theoretical Calculation S4.1 Following the guidance in step S3 (the average d-electron number of the catalyst transition metal is between 5.5 and 8.7), two common bimetallic combinations falling within this range were selected for DFT calculation verification: Co-Cu (average d-electron number = 8.0) and Cu-Mn (average d-electron number = 6.67). Figure 2 (b) SHAP analysis results show that the number of outermost electrons on the support has the least impact on the degradation rate constant. Therefore, the catalyst support was chosen to be the most commonly used C support in the dataset, and N was doped to enhance the metal content. Carrier interaction.

[0062] The molecular structure of the bimetallic catalyst was constructed based on the average d-electron number of the transition metal and the outermost electron number of the support. A finite-size graphene fragment model was constructed using Gaussian software, and two C atoms were replaced with graphite N atoms, with their positions optimized. Two C atoms were removed from the optimized structure to create two vacancies, and corresponding metal atoms were introduced into these vacancies. The initial distances between the metal atoms and the C and N atoms were set to within 2 Å, and the positions were optimized to obtain the catalyst structure. An optimized PMS structure was placed on top of the optimized catalyst structure, and the PMS adsorption configuration of the catalyst was optimized under the same computational level.

[0063] S4.2 uses Gaussian software to optimize the geometry and vibrational frequencies of the catalytic material before and after oxidant adsorption, based on the B3LYP / def2SVP theoretical framework. GD3BJ dispersion correction is introduced. Cartesian coordinates and the keywords Calcfc and NoEigenTest are used during optimization to ensure convergence stability. The keyword combination for the input file in the optimization and vibrational frequency calculations is opt=(calcfc,noeigentest,cartesian) freq b3lyp def2svpem=gd3bj.

[0064] Based on the stable structure optimized in S4.2, S4.3 uses the B3LYP / 6-311++G(d,p) method combined with GD3BJ dispersion correction to calculate the single-point energy. The calculation formula is as follows:

[0065] in The ground-state total energy of the composite system formed by the catalyst and PMS; The ground-state energy of the catalyst before PMS adsorption is given. ν is the ground state energy of the free PMS molecule.

[0066] The optimized structure diagram of the catalyst after PMS adsorption is shown below. Figure 4 As shown, PMS exhibits different structural configurations after adsorption by the two catalyst materials. In the Co-Cu material, the structure of PMS does not change significantly; however, in the Cu-Mn material, the peroxide bond length of PMS increases from 1.46 Å to 2.89 Å. This indicates that material adsorption may lead to the breaking of the peroxide bonds in PMS, thereby generating sulfate radicals and hydroxyl radicals, which can participate in the antibiotic degradation process.

[0067] S5. Determination of the metal element composition of bimetallic catalysts Based on the adsorption energy of S4, d-orbital projected density of states analysis was performed using Multiwfn software based on the output file of Gaussian software. This clarified the adsorption behavior and activation mechanism between bimetallic d-orbitals and between metal d-orbitals and oxidants, thereby screening out the metal element components of bimetallic catalysts with stronger PMS activation ability.

[0068] Based on the projection density of states of the transition metal 3d orbitals and the 2p orbitals of O atoms on PMS ( Figure 5 It can be seen that the metal d orbitals coincide with the 2p orbitals of the O atoms in PMS, and the positions of the density of states peaks of the two orbitals are close, and they have similar trends in intensity distribution. This indicates that the orbital energy matching between the material and PMS is good, and hybridization is likely to occur to form strong chemical bonds.

[0069] Table 2 shows the adsorption energies of different bimetallic catalyst combinations for PMS and the distance between the d-band center and the Fermi level before and after PMS adsorption. Before adsorption, the distance between the d-band center and the Fermi level of the Cu-Mn metal combination is smaller, and theoretically it has a stronger adsorption effect on PMS. Corresponding to the calculated adsorption energy, the Cu-Mn metal combination was selected for catalyst synthesis in this embodiment.

[0070] Table 2 DFT theoretical calculation results

[0071] Based on the optimal design principles derived from model identification and DFT theoretical calculations, the bimetallic catalyst material selected in this embodiment is Cu-Mn with a metal ratio of 1:2, corresponding to an average d-electron number of 6.67 for the transition metal; N-doped C is selected as the support, corresponding to a support with 4 outermost electrons. Furthermore, based on PDP analysis results, the heat treatment temperature and time for catalyst preparation were determined to be 800℃ and 2h, respectively; regarding reaction conditions, the PMS dosage was set to 1.5mM, and the catalyst dosage was set to 0.25g / L, serving as the integrated initial optimized conditions for achieving efficient antibiotic degradation.

[0072] Example 2 Experimental Verification (1) Catalyst design and synthesis The catalyst precursor was prepared using a solution-gel method. 15 g of urea was dispersed in 50 ml (25 ml deionized water + 25 ml anhydrous ethanol) Cu(NO3)2·3H2O and Mn(NO3)2·4H2O solutions (3 wt%), with 0.449 g and 0.934 g of nitrate added, respectively. Then, 1.608 g of citric acid was added as a complexing agent, and the pH of the solution was adjusted to approximately 6. The solution was heated and stirred in a water bath at 75–80 °C for 2 h. The resulting sol was transferred to a polytetrafluoroethylene crucible and dried in a 90 °C oven for 4 h to obtain a solid product. The solid product was then transferred to a corundum crucible and heated to 800 °C at a rate of 10 °C / min and held for 2 h. The resulting product was then ground in an agate mortar to obtain the final material.

[0073] (2) Degradation experiment Based on the machine learning PDP results and the data distribution in the dataset, the experimental conditions were determined as follows: 25±2℃, pH unchanged, initial antibiotic concentration of 20 mg / L, catalyst dosage of 0.25 g / L, PMS dosage of 1.5 mM, and quencher of 10 mM sodium thiosulfate. The reaction was carried out in a 100 ml beaker. At the beginning of the experiment, a specific amount of catalyst was added and stirred for 30 min to remove the influence of adsorption on the experiment. After 30 min, PMS was added and samples were taken simultaneously as the starting point of the degradation experiment. Samples were taken at 0.5, 1, 1.5, 3, 6, 9, 12, 15, 18, 21, 24, 27, and 30 min after the addition of PMS. After filtration through a 0.22 μm filter membrane, the samples were added to a liquid chromatography vial containing the quencher, and the remaining antibiotic concentration was determined by high performance liquid chromatography.

[0074] The activation of PMS by bimetallic catalyst materials for antibiotic degradation, such as Figure 6 As shown. By Figure 6 It can be seen that Cu-Mn / CN almost completely degraded enrofloxacin within 30 min, with a degradation rate of 99.23%, and the degradation rate constant k obtained by pseudo-first-order kinetic fitting was 0.165 min. -1 When the characteristic variables involved in the above experiments were input into the trained XGBoost model, the corresponding degradation rate was 92.73%. These results sufficiently demonstrate that the bimetallic catalyst synthesized based on machine learning and DFT theoretical calculations can efficiently degrade antibiotics in water, and the catalytic material development path proposed in this invention possesses practicality and great application potential.

[0075] Example 3 A screening device and storage medium for Fenton-like reaction bimetallic catalysts guided by machine learning and DFT theoretical calculations A screening device for Fenton-like bimetallic catalysts based on machine learning and DFT theoretical calculations includes: a data acquisition module, a catalyst design module, a DFT calculation screening module, and a human-computer interaction interface.

[0076] A computer device includes a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of a method for predicting the composition of a Fenton-like bimetallic catalyst material guided by machine learning and DFT theory calculations as described in Example 1.

[0077] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for predicting the composition of a Fenton-like bimetallic catalyst material guided by machine learning and DFT theory calculations as described in Example 1.

[0078] Comparative Example 1: Other Machine Learning Models The XGBoost model in Example 1 was replaced with common models such as K-Nearest Neighbors (KNN), Random Forest (RF), Lightweight Gradient Boosting Machine (LightGBM), and Transformer on the same dataset for comparison. All models used the same training / test set partitioning and optimization process. Performance comparisons are shown in Table 3. The XGBoost model achieved higher performance in R... 2 It performed best in terms of RMSE and MAE, verifying its applicability in this invention.

[0079] Table 3 Comparison of training results of different machine learning models

[0080] Comparative Example 2: Choosing Different Target Variables The target feature variable in Example 1 was replaced by the degradation rate constant k with the degradation rate, and the results are shown in Table 4. The XGBoost model, after training, exhibited low interpretability, large errors, and significant overfitting.

[0081] Table 4 Comparison of different target variables

[0082] Comparative Example 3: Selection of Theoretical Calculation Method The choice of theoretical basis set and functional determines the "physical reality" of the potential energy surface; selecting an appropriate computational level can improve computational efficiency and accuracy. In Example 1, the DFT theoretical computational level was replaced from B3LYP / def2SVP to HF / 3-21G and B3LYP / 6-311G. The results are shown in Table 5. The HF / 3-21G level is generally considered "inaccurate" when quantitative results are required or when describing weak interactions or electron-related effects; under the B3LYP functional condition, def2SVP was chosen instead of 6-311G. This is because def2SVP automatically invokes pseudopotentials for heavy elements such as transition metals, and its convergence behavior is more stable in actual calculations, avoiding the 6-311G problem. The non-convergence problem often occurs due to the inappropriate description of heavy atoms by the all-electron basis set. At the same time, this combination has been extensively verified in catalysis and materials systems, and the results are reliable.

[0083] Table 5 Comparison of DFT theoretical calculation results at different computational levels

Claims

1. A method for screening Fenton-like reaction bimetallic catalysts based on machine learning and DFT theory calculation guidance, characterized in that, include: S1. Obtain data on the degradation of pollutants in water by bimetallic catalysts activating oxidants in Fenton-like reactions; The data information includes: the average number of d electrons in the transition metal of the bimetallic catalyst, the number of outermost electrons in the support of the bimetallic catalyst, the catalyst preparation conditions, and the pollutant degradation reaction conditions. S2. Using the data from S1 as input and the degradation rate constant as the target feature, construct a pollutant degradation rate constant prediction model based on the XGBoost algorithm; S3. Set the expected degradation rate constant k value to obtain the average d electron number of the transition metal in the bimetallic catalyst, the number of outermost electrons of the support, and the catalyst preparation conditions; S4. Construct the molecular structure of the bimetallic catalyst based on the average number of d electrons of the transition metal and the number of outermost electrons of the support, and perform DFT theoretical calculations to obtain the adsorption energy of the bimetallic catalyst with different metal element compositions for the oxidant. S5. Select the metal element composition of the bimetallic catalyst based on the adsorption energy of S4, and determine the preparation conditions of the bimetallic catalyst based on the catalyst preparation conditions obtained in S3.

2. The method of claim 1, wherein, In S1, the catalyst preparation conditions include heat treatment temperature and heat treatment time; And / or, the pollutant degradation reaction conditions include reaction temperature, initial pH value, initial pollutant concentration, catalyst dosage, and oxidant dosage; And / or, the contaminant is an antibiotic; And / or, the oxidant is one of persulfate, perdisulfate or hydrogen peroxide.

3. The method as described in claim 1, characterized in that, In S2, the data information is input after being standardized to a uniform scale space by Z-score. Preferably, the Z-score standardization process uses the StandardScaler tool provided by the scikit-learn library.

4. The method as described in claim 1, characterized in that, In S2, the pollutant degradation rate constant prediction model uses grid search and five-fold cross-validation to optimize hyperparameters. The optimized parameters include: the number of decision trees is set to 50~200, the maximum depth is set to 3~9, the learning rate is set to 0.01~0.2, the subsampling ratio is set to 0.6~1.0, the column sampling ratio is set to 0.6~1.0, the L1 regularization parameter is set to 0~0.5, and the L2 regularization parameter is set to 1~2.

5. The method as described in claim 1, characterized in that, In S3, after obtaining the target feature parameters, the prediction model is further explained by SHAP value analysis, and the contribution of each input feature of S2 to the degradation rate constant k is quantified; and the nonlinear influence of key input features on the degradation rate constant k during catalyst design and degradation experiment is revealed by PDP plot analysis.

6. The method as described in claim 1, characterized in that, In S4, the specific steps for the DFT theoretical calculation are as follows: S4.

1. Using Gaussian software, under the theoretical level of B3LYP / def2SVP, the geometric structure and vibration frequency of the catalytic material before and after adsorption of oxidant were optimized and calculated. GD3BJ dispersion correction was introduced. Cartesian coordinates and the keywords Calcfc and NoEigenTest were used in the optimization process to ensure convergence stability. S4.

2. Based on the stable structure optimized in S4.1, the single-point energy is calculated using the B3LYP / 6-311++G(d,p) method combined with GD3BJ dispersion correction. Preferably, in S4.1, the keyword combination of the input file in the optimization calculation and vibration frequency calculation is opt=(calcfc,noeigentest,cartesian) freq b3lyp def2svp em=gd3bj.

7. The method as described in claim 1, characterized in that, In S5, based on the adsorption energy of S4, the Multiwfn software was used to perform d-orbital projected density of states analysis according to the output file of Gaussian software to clarify the adsorption behavior and activation mechanism between bimetallic d orbitals and between metal d orbitals and oxidant, thereby screening out the metal element components of bimetallic catalysts with stronger oxidant activation ability. Preferably, the metal element composition includes the metal element composition and proportion.

8. A device for predicting the composition of Fenton-like bimetallic catalyst materials based on machine learning and DFT theoretical calculations, characterized in that, include: The data acquisition module is used to acquire data on the degradation of pollutants in water by bimetallic catalysts activating oxidants in Fenton-like reactions. The catalyst design module is used to output the average number of d electrons of the transition metal, the number of outermost electrons of the support, and the catalyst preparation conditions based on the set expected degradation rate constant k value using the trained XGBoost prediction model. The DFT calculation and screening module is used to calculate the adsorption energy of bimetallic catalysts with different metal element compositions, and to determine the metal element composition of the bimetallic catalyst based on the calculation results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, The processor executes a computer program to implement the steps of a method for predicting the composition of a Fenton-like bimetallic catalyst material based on machine learning and DFT theoretical calculations as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of a method for predicting the composition of a Fenton-like bimetallic catalyst material based on machine learning and DFT theoretical calculations as described in any one of claims 1 to 7.

Images