Method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries

Abstract

The present application relates to the technical field of lithium-sulfur battery high-entropy alloy catalyst design, and provides a method for efficiently designing a lithium-sulfur battery high-entropy alloy catalyst. The method first calculates the physical and chemical properties such as adsorption energy, formation energy and electronic structure of the high-entropy alloy catalyst based on the density functional theory; uses the Bayesian optimization algorithm to screen the machine learning model with the best performance; extracts the atomic-scale structure parameters that affect the adsorption performance, and trains the regression prediction model in combination with the physical and chemical property data; when predicting a new catalyst, the characteristic parameters of the new catalyst are input into the regression model, the predicted performance result is output, and the predicted performance result is fed back to the density functional theory calculation step to optimize the subsequent calculation scheme. The method can realize rapid prediction and screening of the adsorption performance of the high-entropy alloy catalyst, improve the design efficiency, and reduce the consumption of calculation resources.

Description

Technical Field

[0001] This invention belongs to the field of new energy materials and electrochemical energy storage technology, specifically involving the theoretical design and screening method of lithium-sulfur battery cathode catalysts, and particularly a rapid screening calculation method for high-entropy alloy catalysts that combines density functional theory and machine learning algorithms. Background Technology

[0002] Lithium-sulfur batteries, as energy storage systems with high specific capacity and high energy density, have attracted widespread attention in recent years. Sulfur cathode materials are abundant and inexpensive, but lithium-sulfur batteries still suffer from problems such as shuttle effect, slow reaction kinetics, and volume expansion in actual operation.

[0003] The dissolution and migration of lithium polysulfide intermediates in the electrolyte can lead to the loss of active materials, negative electrode corrosion, and capacity decay, affecting battery cycle performance. Meanwhile, the sulfur reduction reaction involves multiple electron transfer steps and the formation of the insulating product Li₂S, resulting in high reaction overpotential and poor rate performance. Furthermore, the significant volume change of the sulfur electrode during charge and discharge may lead to electrode structural instability, further impacting cycle life.

[0004] Various catalytic materials have been proposed to alleviate the above problems, including metal oxides, single-atom catalysts, and two-dimensional materials. However, these materials still have certain limitations in terms of conductivity and active site stability, and their design cycle is long and relies on experimental trial and error.

[0005] High-entropy alloys, due to their multi-component composition and synergistic effects, possess excellent structural stability and controllability, showing promising application prospects in the field of catalysis. Density functional theory can be used to analyze their stability, electronic structure, and adsorption performance; however, the compositional space of high-entropy alloys is vast, making efficient screening difficult through theoretical calculations alone. In recent years, machine learning methods have been gradually applied to material performance prediction, but their applicability and accuracy in screening high-entropy alloy catalysts still need improvement, and there is a lack of efficient screening models suitable for the adsorption behavior of polysulfides in lithium-sulfur batteries. Summary of the Invention

[0006] The purpose of this invention is to provide a method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries. By combining density functional theory with machine learning models, the performance of novel catalysts can be rapidly predicted and optimized, thereby improving material design efficiency and reducing computational resource consumption.

[0007] The above-mentioned technical objective of this invention is achieved through the following technical solution: a method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries, comprising the following steps: Density functional theory (DFT) was used to calculate the physicochemical properties of the high-entropy alloy catalyst, including adsorption energy, formation energy, and electronic structure. Based on the physicochemical property data obtained from the DFT calculation above, a Bayesian optimization algorithm was used to train and select the machine learning model with the best performance from a variety of machine learning models. Screening atomic-scale structural parameters that affect adsorption performance; Using the screened atomic-scale structural parameters and the physicochemical property data obtained by DFT calculation as input, a regression prediction model for predicting the adsorption energy of sulfur intermediates in lithium-sulfur batteries is constructed by training the optimal machine learning model. When it is necessary to predict the performance of a new catalyst, the relevant characteristic parameters of the new catalyst are input into the regression prediction model. The characteristic parameters include the screened atomic-scale structural parameters and the physicochemical property data calculated using DFT. The regression prediction model calculates the predictive performance of the new catalyst based on the input feature parameters and its internal algorithm. The prediction performance results are output and fed back to subsequent DFT calculation steps to guide subsequent DFT iteration verification or optimization calculations.

[0008] Furthermore, when performing calculations using density functional theory (DFT), first-principles calculations are used to determine the adsorption energy, formation energy, and electronic density of states of the sulfur intermediate in the high-entropy alloy catalyst under different surface configurations, in order to comprehensively evaluate the physicochemical properties of the catalyst.

[0009] Furthermore, when using the Bayesian optimization algorithm to train and select the best-performing machine learning model from multiple machine learning models, the machine learning models include, but are not limited to, Gradient Boosting Regression (GBR), Random Forest (RFR), Adaptive Boosting Regression (ABR), Gaussian Process Regression (GPR), and Support Vector Regression (SVR). The Bayesian optimization algorithm automatically adjusts the model parameters to optimize model performance based on the physicochemical property data obtained by DFT calculation.

[0010] Furthermore, when screening atomic-scale structural parameters that affect adsorption performance, the specific methods include screening the electronegativity difference, atomic radius difference, and electronic state density near the catalytic site of each component atom as key parameters. These parameters can reflect the influence of the atomic composition and electronic structure of the catalyst on the adsorption performance.

[0011] Furthermore, when constructing the regression prediction model for predicting the adsorption energy of sulfur intermediates in lithium-sulfur batteries, the screened atomic-scale structural parameters and physicochemical property data obtained by DFT calculation are used as input features. The model is trained using the best-performing machine learning model. During the model training process, cross-validation is used to evaluate the model performance to ensure the model's generalization ability and prediction accuracy.

[0012] Furthermore, when the relevant characteristic parameters of the new catalyst are input into the regression prediction model, the characteristic parameters include not only the atomic-scale structural parameters obtained by screening and the physicochemical property data obtained by DFT calculation, but also the chemical composition and crystal structure information of the new catalyst, as well as other factors that may affect the adsorption performance, such as surface morphology and defect density.

[0013] Furthermore, when the regression prediction model calculates the predicted performance of the new catalyst based on the input feature parameters using its internal algorithm, it uses feature importance analysis methods (such as SHAP analysis) to interpret the model prediction results, revealing the degree of influence of each feature on the adsorption performance, and providing guidance for further optimization of the catalyst.

[0014] This invention proposes a method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries by combining density functional theory with machine learning methods, which has the following beneficial effects: By combining density functional theory (DFT) with machine learning methods, the screening and optimization process of high-entropy alloy catalysts was significantly accelerated, reducing the screening cycle of candidate high-entropy alloy catalyst systems for lithium-sulfur batteries from months or even years to days, improving efficiency by several to tens of times, while reducing computational resource consumption by at least 70%, making large-scale screening possible. The designed high-entropy alloy catalysts possess multiple active centers and synergistic catalytic effects, which can significantly improve the rate performance and cycle life of lithium-sulfur battery cathodes. High-performance catalysts with moderate adsorption strength (-3.4 eV to -2.8 eV), such as Cr-HEA and Ru-HEA, were screened through DFT calculations, confirming their optimal catalytic activity. This provides efficient theoretical guidance for the development of lithium-sulfur battery catalysts, solving the problems of low efficiency and high experimental costs in traditional catalyst design, and proposing a moderate adsorption design principle that links adsorption energy and catalytic performance. The adsorption energy of novel high-entropy alloy catalysts (such as Os-HEA and Re-HEA) was predicted using machine learning models, with an error as low as 0.057 after DFT verification. The eV (MAE) demonstrates the model's generalization ability and accuracy. Furthermore, this invention provides scalable ML models and design principles to support the customized development of lithium-sulfur battery catalysts, meet the needs of different application scenarios, and has industrialization potential, providing theoretical support and technical solutions for the industrialization of high-performance lithium-sulfur batteries. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the method of the present invention. Detailed Implementation

[0016] The following detailed description, with reference to specific embodiments, illustrates a method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries according to the present invention. This method includes the following steps: high-throughput screening using DFT calculations, machine learning model construction, regression equation generation, and machine learning model prediction.

[0017] 1. DFT calculation high-throughput screening steps Software and parameter settings: A high-entropy alloy model was constructed using the ATAT software package. DFT calculations were performed using the VASP software package, employing the PAW pseudopotential and PBE functional with a cutoff energy of 450 eV. For energy calculations, a 3×3×1 k-point grid was used in the Brillouin zone; for electronic structure calculations, a 10×10×1 Brillouin zone k-point grid was used. The energy convergence standard was 1×10⁻⁵ eV, and the force convergence standard was 0.05 eV / Å. The van der Waals correction used the DFT-D3 method, with a 15 Å vacuum layer along the z-direction to avoid periodic interactions. A 10×10×10 ų supercell was used for calculating the energies of individual small molecules. Spin polarization was considered in all calculations. Data post-processing was performed using VASPKIT.

[0018] Adsorption energy calculation:

[0019] in, This represents the total energy of the system after the high-entropy alloy catalyst adsorbs sulfides. and These represent the energy of the high-entropy alloy catalyst before adsorption and the energy of the adsorbed sulfide, respectively.

[0020] Free energy calculation:

[0021] in, It is the energy difference between the products and reactants before and after the reaction. It is the difference between the zero-point energies of the products and reactants before and after the reaction, while T represents temperature (generally used to calculate the change of free energy of a substance at room temperature, so T is a constant value of 298.15 K). This represents the entropy change of the products and reactants before and after the reaction. The latter two terms can be obtained using VASPKIT's free energy correction function.

[0022] First, 20 stable bulk structures of high-entropy alloys are constructed, followed by a study on catalyst stability (thermodynamic stability and kinetic stability) to obtain stable high-entropy alloy catalysts. Subsequently, the commonly used (111) crystal plane is adopted for subsequent research. S8 cannot stably exist on the catalyst surface, and this problem is effectively solved by using O2 to form a passivation layer on the catalyst surface. By calculating the adsorption energy of the catalyst for lithium polysulfide and comparing it with other materials, its effect on inhibiting the shuttle effect of lithium polysulfide is explored. Then, the adsorption situation between the catalyst and sulfide adsorption intermediates is studied in depth, and the electron transfer between the catalyst substrate and the sulfide intermediates is investigated by means of differential charge density and Bader charge calculation. In addition, through the analysis of the bond length of the Li-S bond and the calculation using COHP, the bonding strength of the Li-S bond is quantitatively calculated, and the correlation between the length and bonding strength of the Li-S bond and the adsorption energy is explored. Furthermore, by dividing the adsorption strength, the catalysts are classified into three categories: strong adsorption, medium adsorption, and weak adsorption. The classification criteria are: strong adsorption (Eads<-3.4 eV); medium adsorption (-3.4 eV<Eads<-2.8 eV); weak adsorption (Eads>-2.8 eV). Finally, through the calculation of the Gibbs free energy of SRR, high-performance high-entropy alloy catalysts Cr-HEA, Ru-HEA, and Ti-HEA are screened out. It is found that they are all catalysts with medium adsorption strength, leading to the conclusion that high-entropy alloy catalysts with medium adsorption strength are the most suitable as catalysts for the sulfur reduction reaction at the positive electrode of Li-S batteries, and they exhibit the best catalytic activity. The medium adsorption design principle for catalysts is proposed in combination with the Sabatier principle.

[0023] 2. Machine Learning Model Construction Data source: DFT calculation results, including HEA adsorption energy data and Li-S bond length data.

[0024] Feature engineering: Global features include 12 average features such as metal electronegativity (χ), d-band center (ε_d), and atomic radius (R). Local features include adsorption site coordination number (CN), Li-S bond length (L_LiS), and sulfur atom number (N_S).

[0025] Model training: Scikit-learn framework, 5-fold cross-validation, and Bayesian search is adopted for hyperparameter optimization.

[0026] Fifteen features were selected through feature engineering, and further categorized into three types based on the characteristics of the high-entropy alloy catalyst. Each type of feature contains 15 features. The first type consists of the feature values ​​of the unique metal element X in the FeCoNiVX pentagonal high-entropy alloy catalyst, with the same feature symbols as before. The second type consists of the average feature values ​​of all elements in the high-entropy alloy catalyst, with the feature symbols appended to the original symbols (_avg1). The third type consists of the average feature values ​​of all metal atoms within a 5 Å range of the adsorption site, with the feature symbols appended to the original symbols (avg_). Each feature type was trained and predicted using eight machine learning models: ABR, GBR, KNR, KRR, RFR, RR, GPR, and SVR. The machine learning model and feature type with the best predictive performance were selected. After two rounds of parameter adjustment and optimization, model training, and prediction, the best-performing machine learning model was ultimately selected as the GBR model (R²: 0.857, MAE: 0.396). This model uses the second type of features as input, so the second type of features and the GBR model were chosen for subsequent analysis. Using SHAP analysis, the top ten features in terms of feature importance scores for the GBR model were calculated. They are: N_S, Hof_avg1, L_LiS, R_avg1, ed_avg1, M_avg1, CN, Hxf_avg1, IE_avg1, and Eea_avg1. Combined with the Pearson correlation coefficient heatmap, it was found that N_S is the feature with the highest correlation and importance to adsorption energy.

[0027] 3. Regression equations and machine learning model predictions A regression equation was successfully fitted using the SISSO algorithm and validated using DFT calculations. The adsorption energies of lithium-sulfur compounds calculated by DFT and those calculated by the SISSO regression equation showed excellent agreement, with an R² of 0.892 and a MAE of 0.324, demonstrating the effectiveness of the regression equation in predicting the performance of Li-S battery catalysts. Finally, a machine learning model was used to predict the adsorption energies of four new high-entropy alloy catalysts: Os-HEA, Re-HEA, Tc-HEA, and W-HEA. Validation using DFT calculations showed that the machine learning model's predictions were highly accurate, with an R² of 0.992 and an MAE of only 0.057, demonstrating the generalization ability of the trained machine learning model.

[0028] In summary, this embodiment systematically predicts and optimizes the adsorption performance of high-entropy alloy catalysts for lithium-sulfur batteries by combining high-throughput density functional theory (DFT) calculations with a Bayesian optimization-driven machine learning model. Specifically, it not only deeply calculated the adsorption behavior and electronic structure evolution of lithium polysulfide intermediates on various high-entropy alloy surfaces, but also accurately identified key structures and electronic features affecting adsorption performance through regression prediction models. Furthermore, it constructed a clear mapping relationship between structure and performance based on feature importance analysis (such as SHAP), achieving quantitative and interpretable catalyst performance evaluation. Moreover, the model prediction results can be fed back to the DFT calculation module, forming a closed-loop path between performance evaluation and structural optimization, significantly improving catalyst design efficiency and intelligence.

[0029] This invention effectively overcomes the core challenges in the development of lithium-sulfur battery catalysts in existing technologies, such as long screening cycles, low performance prediction accuracy, and insufficient basis for structural control, achieving large-scale, high-precision screening of high-entropy alloy catalysts. Compared with traditional catalyst design methods that rely on experience or trial and error, this invention significantly reduces computational resources and time consumption while ensuring prediction accuracy, promoting the transformation of lithium-sulfur battery catalyst design from experience-driven to data-driven intelligent design. In particular, the "medium adsorption strength catalyst screening mechanism" proposed in this invention not only conforms to the Sabatier principle but also provides a reliable theoretical basis and practical guidance for the targeted development of subsequent high-efficiency catalysts. Therefore, this theoretical calculation method has significant technical advantages and broad application prospects in improving the screening efficiency of lithium-sulfur battery catalysts, accelerating the research and development of new materials, and expanding the application of high-entropy alloys in the energy storage field.

[0030] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries, characterized in that, Includes the following steps: DFT calculations were performed to obtain the physicochemical properties of the high-entropy alloy catalyst, including adsorption energy, formation energy, and electronic structure. Based on physicochemical property data, a Bayesian optimization algorithm was used to train and select the best-performing machine learning model from a variety of machine learning models. Screening atomic-scale structural parameters that affect adsorption performance; Using atomic-scale structural parameters and physicochemical property data as input, a regression prediction model for predicting the adsorption energy of sulfur intermediates in lithium-sulfur batteries is constructed by training the optimal machine learning model. When it is necessary to predict the performance of a new catalyst, the relevant characteristic parameters of the new catalyst are input into the regression prediction model. The characteristic parameters include atomic-scale structural parameters and physicochemical property data. The regression prediction model calculates the predicted performance of the new catalyst based on the input feature parameters and its internal algorithm. The prediction performance results are output and fed back to subsequent DFT calculation steps to guide subsequent DFT iteration verification or optimization calculations.

2. The method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries according to claim 1, characterized in that: DFT calculations use first-principles calculations to determine the adsorption energy, formation energy, and electronic density of states of sulfur intermediates in high-entropy alloy catalysts under different surface configurations.

3. The method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries according to claim 1, characterized in that: Bayesian optimization algorithms are used to select the best-performing machine learning model from gradient boosting regression, random forest regression, adaptive boosting regression, Gaussian process regression, and support vector regression.

4. The method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries according to claim 1, characterized in that: Atomic-scale structural parameters include the electronegativity difference of component atoms, the atomic radius difference, and the electronic state density of catalytic sites.

5. The method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries according to claim 1, characterized in that: The regression prediction model takes the selected atomic-scale structural parameters and physicochemical properties as input and is trained and evaluated using cross-validation.

6. The method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries according to claim 1, characterized in that: The feature parameters input into the regression prediction model also include information on the chemical composition and crystal structure of the catalyst.

7. The method for efficiently designing high-entropy alloy catalysts for lithium-sulfur batteries according to claim 1, characterized in that: The regression prediction model uses feature importance analysis to interpret the prediction results and determine the degree of influence of each feature on the adsorption performance.

Images