A high-entropy alloy catalyst component optimization method

Abstract

The application provides a high-entropy alloy catalyst component optimization method, and the optimization method is: selecting at least five elements from a set composed of elements with catalytic activity, obtaining a plurality of catalyst components, calculating the performance values of the catalyst components by using a density functional theory, and constructing an initial data set of 'component-performance'; the initial data set is arranged into a structured data set by using an RS-CM method to couple regional features and site features; feature screening is performed through a Pearson coefficient; an optimal model is selected; the optimal model selected is used to perform performance prediction on data in a test set, and an optimal component is selected. The application uses machine learning technology in combination with RS-CM features, optimizes the design of a high-entropy alloy catalyst by more accurately describing the local environment of an active site, and then accurately predicts and analyzes the performance of a high-entropy alloy component, so that the catalytic performance of a hydrogen evolution reaction is improved.

Description

Technical Field

[0001] This invention relates to the field of high-entropy alloy composition optimization, specifically a method for optimizing the composition of high-entropy alloy catalysts. Background Technology

[0002] Hydrogen evolution reaction (HER), as a crucial process for renewable energy conversion, is receiving widespread attention. HER has important applications in water electrolysis for hydrogen production, fuel cells, and hydrogen storage, and the choice of catalyst is critical to reaction efficiency. Currently, platinum-based catalysts are widely used in HER due to their excellent catalytic activity, but their high cost and scarcity limit their feasibility for large-scale applications. Therefore, developing low-cost, high-efficiency alternative catalysts has become a research hotspot.

[0003] High-entropy alloys (HEAs) are a class of alloys composed of five or more elements in similar molar proportions, exhibiting excellent mechanical, physical, and chemical properties. Due to their complex composition and high-entropy effect, HEAs demonstrate good performance in catalytic reactions. The compositional space of HEAs is vast, and the complex interactions between different elements increase the challenges of alloy design. In traditional experimental methods, optimizing alloy performance typically requires a large amount of experimental data and time, resulting in low efficiency. With the development of machine learning (ML) technology, machine learning-based materials design methods have gradually become an effective means to solve this problem. Through machine learning models, the relationship between alloy composition and performance can be extracted from massive amounts of data, thereby rapidly predicting and optimizing alloy performance.

[0004] However, existing machine learning methods still face many challenges when processing complex, high-dimensional data of high-entropy alloys. In particular, how to accurately capture the interactions between different regions and specific sites in the alloy remains an unsolved problem. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for optimizing the composition of high-entropy alloy catalysts. This invention utilizes machine learning techniques combined with Region-Site Feature Coupling Method (RS-CM) features to optimize the design of high-entropy alloy catalysts by more accurately describing the local environment of active sites. This allows for precise prediction and analysis of the performance of high-entropy alloy components, aiming to improve the catalytic performance of the hydrogen evolution reaction (HER). By introducing RS-CM features, this invention can more accurately capture the interactions between different regions and specific sites within the alloy, thus providing new ideas and methods for developing efficient and economical catalysts.

[0006] To achieve the above objectives, the specific solution adopted by the present invention is as follows:

[0007] A method for optimizing the composition of a high-entropy alloy catalyst mainly includes the following steps:

[0008] S1. Select at least five elements from a set of catalytically active elements to obtain several catalyst components. Calculate the performance values ​​of each catalyst component using density functional theory to construct an initial dataset of "component-performance".

[0009] S2. Use the RS-CM method to organize the initial dataset into a structured dataset;

[0010] S21. Determine the adsorption sites of each component, and use several physical and chemical properties as features to structurally represent the adsorption sites of each component. Divide the adsorption sites into different regions according to the distance between the adsorption sites and the adsorbate.

[0011] S22. Calculate the regional and site features of different regions using several features;

[0012] S23. Use the RS-CM method to couple regional features and site features to organize the initial dataset into a structured dataset;

[0013] S3. Use the Pearson coefficient to filter features, select features that are strongly correlated with the performance values ​​calculated in step S1, and remove redundant features to obtain the filtered dataset. Divide the filtered dataset into training set and test set.

[0014] S4. Select at least two suitable machine learning models, train them using the training set, and use Bayesian optimization to find the optimal parameters for at least two machine learning models. After determining the optimal parameters for at least two machine learning models, use 10x cross-validation to evaluate the stability and generalization ability of at least two machine learning models. Then, use root mean square error and R² to evaluate the stability and generalization ability of at least two machine learning models. 2 The performance of different machine learning models is evaluated and compared to select the best model.

[0015] S5. Use the selected optimal model to predict the performance of the data in the test set and select the optimal component.

[0016] Furthermore, in step S21, the adsorption sites are divided into three different regions based on the distance between each adsorption site and the adsorbate:

[0017] Region 1: Composed of atoms closest to the adsorbed substance;

[0018] Region 2: Composed of atoms in the same plane that are the second closest to the adsorbate;

[0019] Region 3: Composed of atoms closest to the lower layer of the adsorbed substance.

[0020] Further, in step S21, the adsorption sites of each component are structurally represented using the following features: atomic radius R, atomic number AN, d electron number dN, period number PN, group number GN, ionic radius IR, covalent radius CR, lattice constant of pure metal PC, first ionization energy FE, electronegativity EN, electron affinity EA, hydrogen adsorption energy of pure metal *H, d-band center εd of pure metal, p-band center εp of pure metal, s-band center εs of pure metal, and work function WF of pure metal.

[0021] Furthermore, in step S4, the machine learning model is a random forest, support vector machine, gradient boosting regression, or XGBoost regression.

[0022] Furthermore, in step S4, for each machine learning model, the optimal parameters of the model are found by performing 500 Bayesian optimizations, and the stability and generalization ability of the model are evaluated by using tenfold cross-validation.

[0023] Beneficial effects:

[0024] (1) This invention utilizes machine learning techniques combined with RS-CM features to optimize the design of high-entropy alloy catalysts by more accurately describing the local environment of active sites, thereby enabling precise prediction and analysis of the performance of high-entropy alloy components, aiming to improve the catalytic performance of the hydrogen evolution reaction. By introducing RS-CM features, this invention can more accurately capture the interactions between different regions and specific sites in the alloy, thus providing new ideas and methods for developing efficient and economical catalysts.

[0025] (2) Compared with existing catalyst development processes, this invention uses a machine learning model built on a large amount of data and density functional theory calculations to design a catalyst with high HER performance, which greatly saves manpower, material resources and time in the material synthesis and development process. Attached Figure Description

[0026] Figure 1 This is a logical framework diagram of the present invention.

[0027] Figure 2 Performance evaluation graph of the machine learning model for predicting HER performance of HEA designed for Example 1.

[0028] Figure 3 The HER performance test curve of the HEA designed for Example 1.

[0029] Figure 4 A comparison chart of the HER performance of the HEA designed for Example 1 with commercial Pt / C and other catalysts.

[0030] Figure 5 Impedance comparison diagram of HEA designed for Example 1 with commercial Pt / C and other catalysts.

[0031] Figure 6 Stability diagram of the HEA designed for Example 1. Detailed Implementation

[0032] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0033] This invention aims to develop a high-entropy alloy catalyst composition design method to achieve efficient prediction and optimization of catalyst hydrogen adsorption performance. By comprehensively considering regional and site characteristics and combining density functional theory (DFT) calculations, this invention not only improves the accuracy of hydrogen adsorption performance prediction but also accelerates the development process of high-entropy alloys. The data-driven approach employed in this process significantly reduces the time and cost required for experimental verification, thus providing important guidance for the synthesis and application of practical alloys. The steps are described in detail below.

[0034] S1. Obtain the initial dataset of "component-HER performance" through theoretical calculations.

[0035] Select at least five elements from a set of catalytically active elements to obtain several catalyst components. Calculate the performance values ​​of each catalyst component using density functional theory (DFT) to construct an initial "component-performance" dataset.

[0036] S2. Use the RS-CM method to organize the initial dataset into a structured dataset.

[0037] S21. Determine the adsorption sites of each component. Use several physical and chemical properties as features to structurally represent the adsorption sites of each component. In order to more accurately describe the adsorption sites of HEA of different components, divide the adsorption sites into different regions according to the distance between the adsorption sites and the adsorbate.

[0038] This division allows for a more detailed consideration of atomic interactions and environmental influences in different regions. Specifically, Region 1 consists of atoms closest to the adsorbate; due to their proximity, these atoms significantly influence the electronic structure and reactivity of the adsorbate. Region 2 consists of atoms on the same plane closest to the adsorbate; these mid-range adsorption sites play an auxiliary role in the catalytic process, potentially modulating overall catalytic performance by affecting the electron density and reactivity of adjacent atoms. Optimization of these sites may help improve the alloy's selectivity and stability. Region 3 consists of atoms closest to the lower layer of the adsorbate; in this region, the interaction between the adsorption sites and the adsorbate is weak, primarily dominated by indirect effects caused by changes in the local environment. Although these sites have a relatively small direct impact on the catalytic reaction, they can still play a role by influencing the overall structure and stability of the alloy.

[0039] S22. Calculate the regional and site features of different regions using several features.

[0040] Specifically, regional feature representation

[0041]

[0042] Among them, R i Represents the regional characteristics of region i, n i F represents the number of atoms in region i. ij Let be the characteristic of the j-th atom in region i.

[0043] Site feature representation

[0044] S ij =F ij

[0045] Among them, F ij This represents the characteristic of the j-th atom in the i-th region;

[0046] S23. Use the RS-CM method to couple regional features and site features to organize the initial dataset into a structured dataset.

[0047] The core of the RS-CM method is the coupling of regional and site features to accurately capture local and global environmental information of adsorption sites. This coupling considers not only the direct interaction between atoms and sites within a region, but also the spatial and electronic interactions between the region and the site.

[0048] S3, Feature Filtering

[0049] Features are filtered using the Pearson coefficient. The results of the Pearson coefficient help to understand the importance of features and whether there are redundant features. Removing redundant features not only improves model performance but also enhances model interpretability, which is particularly important for understanding the model's predictive mechanism.

[0050] S4. Select at least two suitable machine learning models, train them using the training set, and use Bayesian optimization to find the optimal parameters for at least two machine learning models. After determining the optimal parameters for at least two machine learning models, use 10x cross-validation to evaluate the stability and generalization ability of at least two machine learning models. Then, use root mean square error and R² to evaluate the stability and generalization ability of at least two machine learning models. 2 The score evaluation compares the performance of different machine learning models and selects the best model.

[0051] Choose appropriate machine learning algorithms (such as random forest, support vector machine, gradient boosting regression, XGBoost regression, etc.) to build a prediction model targeting regional and site features. Select the root mean square error (RMSE) and coefficient of determination (R²). 2 () as an evaluation indicator.

[0052] For each model, the optimal hyperparameters are found using 500 Bayesian optimizations. The model's generalization ability is evaluated using 10x cross-validation.

[0053] The definition of RMSE is:

[0054]

[0055] Among them, y i It is the i-th observation (actual value). It is the i-th predicted value, and n is the number of samples.

[0056] The smaller the RMSE value, the better the model's prediction performance and the smaller the difference between the predicted and actual values.

[0057] R 2 The defining formula is:

[0058]

[0059] in, R is the average of the actual values. 2 It is a relative metric, suitable for comparing the performance of different models.

[0060] S5. Use the selected optimal model to predict the performance of the data in the test set and select the optimal component.

[0061] A trained model was used to predict the performance of high-entropy alloys with different combinations, and HEA with high HER performance was selected. Experimental verification was conducted on the selected optimal alloy composition to evaluate its catalytic performance. HER performance data of the catalyst were obtained through electrochemical testing and other methods to further verify the accuracy of the computational model.

[0062] Example 1

[0063] This embodiment discloses a method for optimizing the composition of a high-entropy alloy catalyst. Please refer to [link / reference]. Figure 1 The main steps include the following:

[0064] S1. Construct an initial dataset with a one-to-one correspondence between "composition-property" in high-entropy alloys.

[0065] Specifically, a broad review of relevant literature was conducted to understand the current research progress of high-entropy alloys and their experimental data on hydrogen adsorption performance. This data included the chemical composition of different elemental combinations, the physicochemical properties of the alloys, and their known catalytic properties. The research subjects were determined, selecting elements with potential catalytic activity such as iron (Fe), cobalt (Co), nickel (Ni), chromium (Cr), molybdenum (Mo), manganese (Mn), palladium (Pd), ruthenium (Ru), and rhodium (Rh). To save costs, the three elements FeCoNi were fixed, resulting in a high-entropy alloy combination of FeCoNiXY, with a total of 15 combinations. Performance values ​​for each combination were calculated using DFT, constructing an initial dataset corresponding to "composition-performance." This initial dataset provided a solid foundation for the subsequent training and prediction of machine learning models, and could be used to explore the relationship between alloy composition and electrocatalytic performance, and to screen for alloy combinations with optimal HER performance.

[0066] To ensure a more comprehensive distribution of the dataset, systematic sampling was performed for each combination, guaranteeing coverage of diverse situations and features. This strategy not only improves the representativeness of the data but also enhances the model's robustness under different conditions. In this way, we are able to better capture underlying trends and patterns, thereby improving the model's predictive performance.

[0067] S2. Use the RS-CM method to organize the initial dataset into a structured dataset to ensure a consistent data format, facilitating subsequent analysis and modeling.

[0068] S21. Determine the adsorption sites of each component, and structurally represent the adsorption sites of each component using several physical and chemical properties as characteristics. Divide the adsorption sites into different regions according to the distance between the adsorption sites and the adsorbate.

[0069] Specifically, to accurately describe the properties of the adsorption sites, 16 physical and chemical properties were selected as characteristics to characterize each adsorption site. These characteristics mainly originate from the electronic structure, geometric properties, and interaction forces between the material and the adsorbate, and they can provide rich information for subsequent prediction of catalytic performance. Specifically, the selected characteristics include:

[0070] R (atomic radius), AN (atomic number), dN (d electron number), PN (period number), GN (group number), IR (ionic radius), CR (covalent radius), and PC (lattice constant of pure metals) are characteristics that describe the intrinsic properties of atoms.

[0071] FE (first ionization energy), EN (electronegativity), and EA (electron affinity) are characteristics that describe the chemical activity and reactivity of elements.

[0072] *H (pure metal hydrogen adsorption energy), εd (pure metal d-band center), εp (pure metal p-band center), εs (pure metal s-band center), and WF (pure metal work function) are directly related to the adsorption behavior in catalytic reactions.

[0073] To more accurately describe the adsorption sites of different alloys, the adsorption sites are divided into three regions based on the distance between the adsorption site and the adsorbate. This partitioning method allows for a deeper exploration of the influence of different regions on electrocatalytic performance. This method effectively combines the regional and site characteristics of the adsorption sites, enabling a clearer analysis of the characteristics of the adsorption sites and their behavior during the catalytic process. Region 1: Composed of atoms closest to the adsorbate. Due to their proximity to the adsorbate, these atoms significantly influence the electronic structure and reactivity of the adsorbate. In catalytic reactions, the electronic environment of the adsorption site determines the adsorption strength and reaction pathway of the reactants; therefore, atoms in Region 1 are typically the most active part of the catalytic reaction. These atoms not only interact directly with the adsorbate but also influence the desorption and conversion of the adsorbate by altering the local electron density. In high-entropy alloys, atoms in Region 1 are usually located on or very close to the alloy surface, playing a dominant role in reactivity and catalytic activity. For HER, the adsorption sites in Region 1 can optimize catalytic performance by adjusting the adsorption energy of hydrogen atoms. For example, excessively strong adsorption energy may lead to excessive hydrogen adsorption, thereby hindering the hydrogen desorption process; while excessively weak adsorption energy may lead to unstable hydrogen adsorption, reducing reaction efficiency.

[0074] Region 2: Composed of atoms in the same plane closest to the adsorbate. These atoms are located relatively far from the adsorbate, resulting in weaker interactions, but they can still significantly influence catalytic performance by affecting local electron density and interatomic interactions. Mid-distance adsorption sites play an auxiliary role in catalytic reactions, primarily by influencing the electron density and reactivity of surrounding atoms, thereby modulating overall catalytic performance. Atoms in Region 2 can adjust the electronic structure of adjacent atoms, causing the catalyst to exhibit different activities and selectivities under different reaction conditions. For example, atoms in Region 2 may improve or decrease overall catalytic performance by affecting the electronic environment of the catalyst surface atoms. In the hydrogen evolution reaction, atoms in Region 2 may improve the alloy's selectivity and stability by influencing the desorption process of hydrogen molecules. Region 2 sites in the alloy can modulate the reaction pathway and improve the catalyst's performance under different reaction conditions.

[0075] Region 3: Composed of atoms closest to the adsorbate in the lower layer, these atoms are relatively far from the adsorbate and typically exhibit weak direct interactions. In this region, the interaction between the adsorption sites and the adsorbate is weak, dominated primarily by indirect effects caused by changes in the local environment. Although these sites have a relatively small direct impact on the catalytic reaction, they can still play a significant role by influencing the overall structure and stability of the alloy. Atoms in Region 3 are located in the subsurface layer of the alloy, and while they have less direct interaction with the catalytic reaction, they can indirectly affect catalytic performance by influencing the stability of the overall structure. For example, sites in Region 3 may play a role in long-term reaction processes, improving the durability and stability of the alloy. In the HER reaction, the alloy surface may undergo long-term catalytic cycling; sites in Region 3 help the catalyst maintain high performance during the reaction by maintaining the overall structure and stability of the alloy. Although the direct catalytic activity of these sites is low, they are crucial for the long-term stability of the catalyst. Therefore, understanding and optimizing the adsorption sites in these regions is of great significance for extending the catalyst's lifetime.

[0076] By dividing the adsorption sites into three regions, the specific contributions of adsorption sites in different regions to catalytic performance can be more clearly identified and analyzed. This partitioning method not only helps in understanding the catalytic mechanism of alloy materials but also provides more targeted characteristic information for subsequent model construction. Through this method, the adsorption mechanism can be analyzed in depth at the molecular level, thereby optimizing the electrocatalytic performance of high-entropy alloys.

[0077] S22. Calculate the regional and site features of different regions using several features.

[0078] Regional feature representation

[0079]

[0080] Among them, R i Represents the regional characteristics of region i, n i F represents the number of atoms in region i. ij Let be the characteristic of the j-th atom in region i.

[0081] Site feature representation

[0082] S ij =F ij

[0083] Among them, F ij This represents the characteristic of the j-th atom in the i-th region.

[0084] S23. Using the RS-CM method to couple region features and site features, the initial dataset is organized into a structured dataset.

[0085] This meticulous feature coupling provides a more comprehensive theoretical basis for optimizing the catalytic performance of high-entropy alloys. Specifically, the RS-CM method, based on adsorption site partitioning, not only considers atomic interactions within different regions but also enhances the understanding of the adsorption site environment by combining regional and site features. This method allows for more precise capture of complex effects in the catalytic process, including local effects at the atomic level and global effects at the regional level, thus providing a more detailed and reliable predictive basis for optimizing the catalytic performance of high-entropy alloys.

[0086] S3. Feature selection is performed using the Pearson coefficient to identify features with strong correlation to the performance values ​​calculated in step S1, and redundant features are removed to obtain the selected dataset. The selected dataset is then divided into training and test sets.

[0087] To select features that have a significant impact on the target value and avoid excessive redundancy in the dataset, the Pearson correlation coefficient is used as the feature selection criterion. The Pearson correlation coefficient quantifies the linear relationship between two variables, ranging from -1 to 1; a larger absolute value indicates a stronger linear relationship. In feature selection, the Pearson correlation coefficient between each feature and the target value is calculated, and features with a strong correlation to performance values ​​are selected.

[0088] Besides features with weak correlation to the target value, redundant features are also a key aspect to consider in data preprocessing. Redundant features refer to variables that are highly correlated with other features; they provide almost identical information, offer no benefit to the model's predictions, and may even introduce noise and increase computational complexity. To remove redundant features, the correlation between all features is calculated, and a threshold (0.9) is set. When the Pearson correlation coefficient between two features exceeds this threshold, these two features are considered redundant, and the one with the greater impact on performance is selected and retained.

[0089] S4. Select at least two suitable machine learning models, train them using the training set, and use Bayesian optimization to find the optimal parameters for at least two machine learning models. After determining the optimal parameters for at least two machine learning models, use 10x cross-validation to evaluate the stability and generalization ability of at least two machine learning models. Then, use root mean square error and R² to evaluate the stability and generalization ability of at least two machine learning models. 2 The score evaluation compares the performance of different machine learning models and selects the best model.

[0090] Specifically, during the model selection process, various machine learning models were screened, including Support Vector Regression (SVR), Random Forest Regression (RF), Gradient Boosting Regression (GBR), and XGBoost Regression. These models each have their own characteristics; for example, SVR performs well in handling nonlinear relationships, RF is robust to nonlinear data and noise, while GBR and XGBoost, based on the gradient boosting method, are highly effective at modeling complex data. Therefore, comparing multiple models provides a performance evaluation from different perspectives.

[0091] First, a reasonable parameter range was set for each model, and Bayesian optimization was used to find the optimal combination of hyperparameters. The advantage of Bayesian optimization lies in its probabilistic model-based search strategy, which balances exploration and exploitation, quickly finding near-optimal hyperparameter combinations through a small number of iterations. Compared to traditional grid search and random search, Bayesian optimization significantly reduces search time, especially in high-dimensional parameter spaces.

[0092] After determining the optimal parameters for each model, tenfold cross-validation was used to evaluate their performance. Tenfold cross-validation involves dividing the dataset into ten subsets, alternating between each subset as the validation set and the remainder as the training set, for a total of ten validation runs. This provides a comprehensive assessment of the model's generalization ability and stability, avoiding the influence of the randomness of a single split on the results.

[0093] Finally, the average performance metrics of each model in cross-validation are compared, using root mean square error (RMSE) and R0. 2The scores are used to select the best-performing XGBoost regression model for subsequent prediction tasks. The definition of RMSE is: Among them, y i It is the i-th observation (actual value). Here, is the i-th predicted value, and n is the number of samples. The smaller the RMSE value, the better the model's prediction performance, and the smaller the difference between the predicted and actual values ​​(e.g., ...). Figure 2 (The light gray pillars in a and 2b). R 2 The defining formula is: in, R is the average of the actual values. 2 The value is between 0 and 1, such as Figure 2 The dark gray bars in a and 2b represent the regression models most suitable for practical applications, determined through this multi-model screening and evaluation process, thus improving model reliability and accuracy. For example... Figure 2 As shown, XGBoost regression is the best machine learning model.

[0094] S5. Use the optimized XGBoost regression model for prediction, and select the optimal component by combining the DFT calculation data and machine learning prediction data.

[0095] Specifically, in the final prediction and screening stage, the XGBoost regression model selected and optimized in step S4 is used to predict the catalytic performance of the alloy. This process aims to combine DFT (density functional theory) calculations with prediction data from machine learning models to comprehensively analyze and select the optimal alloy composition.

[0096] First, the features of all candidate alloy components are input into a trained XGBoost regression model for prediction. Thanks to its excellent performance demonstrated in the initial training and cross-validation, the XGBoost regression model can accurately predict the catalytic performance of different alloys under specific conditions.

[0097] By analyzing machine learning predictions and DFT calculations, a more comprehensive assessment of alloy properties can be obtained. DFT calculations offer high theoretical accuracy, particularly advantageous in analyzing specific interatomic electronic structures and microscopic reaction mechanisms. The introduction of the XGBoost regression model significantly improves the breadth and efficiency of predictions, enabling the rapid screening of potential high-performance combinations.

[0098] By combining the rapid predictive capabilities of the XGBoost regression model with the theoretical verification through DFT calculations, the optimal high-entropy alloy composition with excellent HER performance can be efficiently identified and selected. This achieves an organic integration of experimental and computational research, providing guidance for further experimental verification and practical applications.

[0099] Experimental Synthesis of S6 and FeCoNiMoPd

[0100] Specifically, 0.02 mmol of FeCl3·6H2O, 0.02 mmol of NiCl2·6H2O, 0.02 mmol of palladium chloride, 0.02 mmol of molybdenum acetylacetonate, and 0.02 mmol of palladium chloride were accurately weighed and dissolved in 20 ml of anhydrous ethanol to form a homogeneous solution. This solution was then sonicated for 1 hour. This step helps to accelerate the dissolution process and ensures that the metal precursors are fully mixed in the solution, achieving a uniform precursor dispersion.

[0101] After sonication, 200 mg of carbon matrix material was added to the metal solution. This carbon matrix provides a high specific surface area and good conductivity, thereby improving the performance of the catalytic material. The mixture was then stirred overnight at room temperature with a magnetic stirrer to enhance the bonding strength between the precursor and the carbon matrix. The stirred mixture was then freeze-dried for 24 hours. The freeze-drying process, by rapidly freezing the sample at low temperature and then sublimating the solvent under vacuum, avoided pore structure collapse or particle aggregation caused by liquid solvent evaporation, ensuring the pore structure and uniformity of the sample. Next, the dried sample was transferred to a tube furnace and heated to 700 °C at a heating rate of 10 °C / min, and held at this temperature for 1 hour. After the holding period, the sample was cooled to room temperature with the furnace.

[0102] After cooling, the resulting sample was collected and named FeCoNiMoPd-HEA for subsequent characterization and performance testing. This preparation method ensures that the synthesized material possesses excellent metal dispersion and structural stability, thereby enhancing its application potential in fields such as electrocatalysis.

[0103] Performance testing of S7 and FeCoNiMoPd-HEA

[0104] Specifically, in this embodiment, the electrochemical performance of the three-electrode battery system was comprehensively evaluated at room temperature using an electrochemical workstation (CHI660E, Shanghai, China). The three-electrode system includes: a catalyst-coated carbon cloth as the working electrode (loading 1 mg / cm³). 2 A platinum wire was used as the counter electrode, and a mercurous sulfate electrode was used as the reference electrode. The electrolyte for the HER experiment was a 0.5 M H₂SO₄ aqueous solution.

[0105] For electrochemical testing, cyclic voltammetry (CV) was used to activate the catalyst. The test potential range was 0.168 to -0.532 V (relative to the reversible hydrogen electrode, RHE), with a scan rate of 100 mV / s, and 20 cycles were performed to fully activate the catalyst surface. Subsequently, linear sweep voltammetry (LSV) was performed within the same potential range at a scan rate of 5 mV / s to evaluate the HER activity of the electrode. Figure 3 The figure shows the HER performance curve of HEA. Figure 4 This demonstrates that the selected HEA catalyst exhibits superior HER performance compared to other catalysts.

[0106] Electrochemical impedance spectroscopy (EIS) measurements were performed in the frequency range of 100 kHz to 0.01 Hz with an AC voltage of 5 mV amplitude to analyze the charge transfer resistance and other kinetic properties of the electrodes. Furthermore, to evaluate the catalyst durability, measurements were performed at a constant current density of 10 mA / cm². 2 The timing potential curve was used to test its stability and long-term operating performance. For example... Figure 5 The impedance of the selected HEA is lower than that of other catalysts, which proves that it has a strong charge transfer capability. Figure 6 To demonstrate stability, the screened HEA catalyst was tested at 10 mA cm⁻¹. -2 It has a stability of over 2000 hours.

[0107] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention in any way. All equivalent transformations or modifications made in accordance with the essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for optimizing the composition of a high-entropy alloy catalyst, characterized in that, The main steps include the following: S1. Select at least five elements from a set of catalytically active elements to obtain several catalyst components. Calculate the performance values ​​of each catalyst component using density functional theory to construct an initial dataset of "component-performance". The catalyst is a hydrogen evolution reaction catalyst, and catalytic activity refers to the catalytic activity of the hydrogen evolution reaction. S2. Use the RS-CM method to organize the initial dataset into a structured dataset, where RS-CM stands for Region-Site Feature Coupling Method. S21. Determine the adsorption sites of each component, and use several physical and chemical properties as features to structurally represent the adsorption sites of each component. Divide the adsorption sites into different regions according to the distance between the adsorption sites and the adsorbate. S22. Calculate the regional and site features of different regions using several features; S23. Use the RS-CM method to couple regional features and site features to organize the initial dataset into a structured dataset; S3. Use the Pearson coefficient to filter features, select features that are strongly correlated with the performance values ​​calculated in step S1, and remove redundant features to obtain the filtered dataset. Divide the filtered dataset into training set and test set. S4. Select at least two suitable machine learning models, train them using the training set, use Bayesian optimization to find the optimal parameters of at least two machine learning models, after determining the optimal parameters of at least two machine learning models, use 10x cross-validation to evaluate the stability and generalization ability of at least two machine learning models, and then use root mean square error and R² score to evaluate and compare the performance of different machine learning models and select the optimal model. S5. Use the selected optimal model to predict the performance of the data in the test set and select the optimal component.

2. The method for optimizing the composition of a high-entropy alloy catalyst according to claim 1, characterized in that, In step S21, the adsorption sites are divided into three different regions based on the distance between each adsorption site and the adsorbate: Region 1: Composed of atoms closest to the adsorbed substance; Region 2: Composed of atoms in the same plane that are the second closest to the adsorbate; Region 3: Composed of atoms closest to the lower layer of the adsorbed substance.

3. The method for optimizing the composition of a high-entropy alloy catalyst according to claim 1, characterized in that, In step S21, the adsorption sites of each component are structurally represented using the following characteristics: atomic radius R, atomic number AN, d electron number dN, period number PN, group number GN, ionic radius IR, covalent radius CR, lattice constant of pure metal PC, first ionization energy FE, electronegativity EN, electron affinity EA, hydrogen adsorption energy of pure metal *H, d-band center εd of pure metal, p-band center εp of pure metal, s-band center εs of pure metal, and work function WF of pure metal.

4. The method for optimizing the composition of a high-entropy alloy catalyst according to claim 1, characterized in that, In step S4, the machine learning model is random forest, support vector machine, gradient boosting regression, or XGBoost regression.

5. The method for optimizing the composition of a high-entropy alloy catalyst according to claim 1, characterized in that, In step S4, for each machine learning model, the optimal parameters of the model are found by performing 500 Bayesian optimizations, and the stability and generalization ability of the model are evaluated by using tenfold cross-validation.

Images