Method for the autonomous synthesis of high-entropy oxide electrocatalytic materials

Abstract

The application discloses a method for self-synthesis of high-entropy oxide electrocatalytic materials, comprising the following steps: S1, constructing a high-entropy oxide electrocatalytic material database; S2, increasing sample data in the high-entropy oxide electrocatalytic material database; S3, training a multi-module fusion machine learning research and development model for the electrocatalytic material database; S4, receiving a preset target electrocatalytic performance index; S5, verifying and analyzing model simulation analysis results; S6, referring to the analysis results in S5 to trial-produce high-entropy oxides; S7, analyzing the performance of finished high-entropy oxides, optimizing material proportions and preparation process parameters; S8, referring to the analysis results in S7 to trial-produce high-entropy oxides; S9, repeating S7 and S8 until the performance of high-entropy oxides is optimal; and S10, adding process parameters to the high-entropy oxide electrocatalytic material database. The application has the benefits of high research and development efficiency, accurate proportion control, optimized process, and short research and development cycle.

Description

Technical Field

[0001] This invention relates to the field of materials research and development technology, and more specifically to a method for the autonomous synthesis of high-entropy oxide electrocatalytic materials. Background Technology

[0002] With the rapid development of the new energy industry, clean energy conversion and storage technologies have become the core key to solving the energy crisis. Electrocatalysis, as a core technology in clean energy conversion processes such as water electrolysis, fuel cells, and carbon dioxide reduction, directly determines the energy conversion efficiency and industrial application value of its catalytic materials. High-entropy oxides, as a new type of functional material formed by the uniform mixing of multiple main metal oxides, have broken through the performance limitations of traditional single-component and binary composite oxide catalytic materials due to their unique physicochemical properties such as high configurational entropy, lattice distortion, hysteresis diffusion, and cocktail effect.

[0003] Compared to traditional catalytic materials, high-entropy oxides possess significant advantages such as multi-component synergistic regulation, abundant active sites, and strong structural stability. The multiple metal components within them can achieve near-infinite permutations and combinations, providing a broad research platform for structural modification, component regulation, and performance optimization of catalytic materials. Furthermore, the synergistic effect of multiple principal components not only balances the material's fundamental properties such as conductivity, catalytic activity, and structural stability, but also derives novel physicochemical properties not found in traditional single-component materials through the mutual coupling between components. This provides a new approach to developing high-performance, multifunctional electrocatalytic materials, completely breaking down the technical barriers of single-performance and limited regulatory space inherent in traditional catalytic materials.

[0004] Furthermore, traditional materials research and development relies on the conventional "experiment-characterization-analysis" research paradigm, which suffers from drawbacks such as long development cycles, high experimental costs, numerous trial-and-error attempts, and difficulty in performance prediction. It fails to quickly uncover the optimal combination of multiple components in high-entropy materials, making it difficult to fully leverage the performance advantages of multi-component regulation in high-entropy materials. Traditional research and development models struggle to accurately predict the electrocatalytic performance of high-entropy oxides with different components and structures, and cannot efficiently solve the problem of the correlation between configurational entropy and catalytic mechanisms, severely hindering the iterative development speed of high-performance high-entropy electrocatalytic materials. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a method for the autonomous synthesis of high-entropy oxide electrocatalytic materials that features high R&D efficiency, precise control of proportions, optimized processes, and a short R&D cycle.

[0006] According to one aspect of the present invention, a method for the autonomous synthesis of high-entropy oxide electrocatalytic materials is provided, comprising the following steps: S1. Construct a database of high-entropy oxide electrocatalytic materials; S2. Add multi-dimensional sample data to the high-entropy oxide electrocatalytic materials database, including microstructure parameters, preparation process parameters, electrocatalytic performance parameters, atomic simulation data, and scientific research text knowledge data. S3. Train and construct a multi-module integrated machine learning R&D model based on the electrocatalytic material database to establish the mapping relationship between material characteristics, process parameters and electrocatalytic performance; S4. Receive the preset target electrocatalytic performance indicators; S5. Verify the simulation analysis results of the analysis model; S6. Prepare high-entropy oxides by referring to the analysis results in step S5; S7. Analyze the properties of the finished high-entropy oxide and optimize the material ratio and preparation process parameters; S8. Prepare high-entropy oxides by referring to the analysis results in step S7; S9. Repeat steps S7 and S8 until the high-entropy oxide performs optimally. S10. Add the optimized material ratios and preparation process parameters to the high-entropy oxide electrocatalytic materials database.

[0007] In some implementations, the machine learning development model constructed in step S3 includes a material property prediction proxy model. The material property prediction proxy model is trained based on high-entropy oxide sample data within the dataset and fits the correlation mapping relationship between material composition, microstructure and electrocatalytic activity, structural stability and catalytic selectivity.

[0008] In some implementations, the machine learning development model is equipped with an adaptive active learning iterative framework. Based on the experimental feedback data of the candidate materials, the active learning iterative framework continuously updates the model weight parameters, adaptively optimizes the high-entropy oxide composition ratio and preparation process parameters, and dynamically improves the model prediction accuracy and the performance matching degree of the candidate materials.

[0009] In some implementations, the machine learning development model integrates a variant autoencoder (VAE) and a generative adversarial network (GAN). The VAE is used to perform dimensionality reduction, extraction, and standardized characterization of the microstructural features of high-entropy oxides. Based on the characterization features, the GAN intelligently generates multi-component novel high-entropy oxide candidate structures, realizing innovative structural design and system expansion of high-entropy electrocatalytic materials.

[0010] In some implementations, the machine learning development model is configured with a machine learning atomic force field, and atomic-scale simulation calculations of high-entropy oxides are carried out through the machine learning atomic force field. Based on the simulation results, the evolution laws of lattice distortion, component diffusion, and surface active sites of high-entropy oxides are analyzed, and the intrinsic correlation mechanism between the configuration entropy of high-entropy oxides and the electrocatalytic reaction mechanism is established.

[0011] In some implementations, the machine learning development model integrates a deep learning representation module, which is used to collect and analyze atomic-scale microscopic imaging data of high-entropy oxides, intelligently identify and accurately analyze transmission electron microscopy and scanning electron microscopy imaging data, and quantitatively characterize the microscopic morphology, defect structure and multi-component distribution characteristics of the material.

[0012] In some implementations, natural language processing (NLP) technology is integrated into the construction of the dataset. NLP technology is used to perform structured parsing of scientific research literature, patent texts, and experimental reports in the field of high-entropy materials, automatically extracting the related knowledge of material composition, preparation process, and performance characteristics, supplementing and improving the sample dimensions of the dataset, and providing data support for machine learning model training.

[0013] In some implementations, the target electrocatalytic performance indicators include any one or more of the following: water electrolysis catalytic performance indicators, fuel cell catalytic performance indicators, and carbon dioxide reduction catalytic performance indicators. The machine learning development model differentiates and screens suitable high-entropy oxide candidate material systems based on the indicator thresholds for different electrocatalytic application scenarios.

[0014] Compared with existing technologies, this invention has the advantages of high R&D efficiency, precise control of proportions, optimized processes, and short R&D cycle. This invention constructs a multi-dimensional closed-loop data R&D system, integrating experimental data, simulation data, and literature and patent text data. This solves the problems of fragmented data and insufficient sample size in traditional high-entropy catalytic material R&D. Through continuous data iteration and updates, the accuracy of material R&D is constantly improved. By integrating a multi-module machine learning model, including VAE-GAN generation network, atomic force field simulation, deep learning representation, and active learning iterative framework, it can achieve autonomous design of novel high-entropy oxide structures, accurate performance prediction, and in-depth mechanism analysis, completely eliminating the manual trial-and-error R&D mode and shortening the R&D cycle by more than 60%. Through adaptive iterative optimization capabilities, through bidirectional iteration of experimental test feedback and model weight updates, it continuously optimizes material ratios and preparation processes, effectively improving the activity, stability, and selectivity of high-entropy oxides, and solving the problem of low performance control precision in traditional preparation processes. This invention is adaptable to various electrocatalytic scenarios such as water electrolysis, fuel cells, and carbon dioxide reduction. It can screen material systems according to the performance threshold differences of different scenarios, exhibiting strong versatility and practicality, and providing intelligent technical support for the industrialization R&D of high-performance high-entropy electrocatalytic materials. Detailed Implementation

[0015] The present invention will now be described in detail with reference to various embodiments. However, it should be noted that these embodiments are not intended to limit the present invention. Equivalent transformations or substitutions in function, method or structure made by those skilled in the art based on these embodiments are all within the protection scope of the present invention.

[0016] In the description of this invention, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the terms according to the specific circumstances.

[0017] The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to the present invention includes the following steps: S1. Construct a database of high-entropy oxide electrocatalytic materials; S2. Add multi-dimensional sample data to the high-entropy oxide electrocatalytic materials database, including microstructure parameters, preparation process parameters, electrocatalytic performance parameters, atomic simulation data, and scientific research text knowledge data. S3. Train and construct a multi-module integrated machine learning R&D model based on the electrocatalytic material database to establish the mapping relationship between material characteristics, process parameters and electrocatalytic performance; S4. Receive the preset target electrocatalytic performance indicators; S5. Verify the simulation analysis results of the analysis model; S6. Prepare high-entropy oxides by referring to the analysis results in step S5; S7. Analyze the properties of the finished high-entropy oxide and optimize the material ratio and preparation process parameters; S8. Prepare high-entropy oxides by referring to the analysis results in step S7; S9. Repeat steps S7 and S8 until the high-entropy oxide performs optimally. S10. Add the optimized material ratios and preparation process parameters to the high-entropy oxide electrocatalytic materials database.

[0018] The machine learning R&D model constructed in step S3 includes a material property prediction proxy model. The material property prediction proxy model is trained based on high-entropy oxide sample data in the dataset and fits the correlation mapping relationship between material composition, microstructure and electrocatalytic activity, structural stability and catalytic selectivity.

[0019] The machine learning development model is equipped with an adaptive active learning iterative framework. Based on the experimental feedback data of candidate materials, the active learning iterative framework continuously updates the model weight parameters, adaptively optimizes the high-entropy oxide composition ratio and preparation process parameters, and dynamically improves the model prediction accuracy and the performance matching degree of candidate materials.

[0020] The machine learning research model integrates the Mutation Autoencoder (VAE) and the Generative Adversarial Network (GAN). The VAE is used to reduce the dimensionality, extract and standardize the microstructural features of high-entropy oxides. Based on the characterization features, the GAN intelligently generates multi-component novel high-entropy oxide candidate structures, realizing innovative structural design and system expansion of high-entropy electrocatalytic materials.

[0021] The machine learning research and development model is configured with a machine learning atomic force field, and atomic-scale simulation calculations of high-entropy oxides are carried out through the machine learning atomic force field. Based on the simulation results, the evolution law of lattice distortion, component diffusion and surface active site of high-entropy oxides is analyzed, and the intrinsic correlation mechanism between the configuration entropy of high-entropy oxides and the electrocatalytic reaction mechanism is established.

[0022] The machine learning R&D model integrates a deep learning representation module, which is used to collect and analyze atomic-scale microscopic imaging data of high-entropy oxides, and to intelligently identify and accurately analyze transmission electron microscopy and scanning electron microscopy imaging data to quantitatively characterize the microscopic morphology, defect structure and multi-component distribution characteristics of the material.

[0023] Natural language processing (NLP) technology is integrated into the construction of the dataset. Through NLP, scientific research literature, patent texts, and experimental reports in the field of high-entropy materials are structured and parsed to automatically extract the related knowledge of material composition, preparation process, and performance characteristics, thereby supplementing and improving the sample dimensions of the dataset and providing data support for machine learning model training.

[0024] The target electrocatalytic performance indicators include any one or more of the following: water electrolysis catalytic performance indicators, fuel cell catalytic performance indicators, and carbon dioxide reduction catalytic performance indicators. The machine learning development model is used to differentiate and screen suitable high-entropy oxide candidate material systems based on the indicator thresholds for different electrocatalytic application scenarios.

[0025] Example 1: In-house synthesis of high-entropy oxide electrocatalysts adapted for oxygen evolution reaction (OER) in water electrolysis S1. Build a dedicated database for high-entropy oxide electrocatalytic materials, establish a standardized framework for data classification and storage, feature annotation, and data retrieval, and initialize the database storage structure.

[0026] S2. Multidimensional Expansion of the Dataset: Relying on natural language processing technology, we crawled and analyzed literature, patents, and experimental reports in the field of high-entropy oxide water electrolysis catalysis over the past 10 years. We automatically extracted process parameters such as metal composition ratio, sintering temperature, holding time, and precursor concentration, as well as microstructure parameters such as lattice constant, defect density, and pore size distribution, and electrocatalytic performance parameters such as overpotential, Tafel slope, and cycle stability. At the same time, we combined machine learning atomic force fields to complete atomic-scale simulations of Ni-Co-Fe-Mn-Cu pentagonal high-entropy oxides, supplementing simulation data such as lattice distortion and active site distribution, and constructing a multidimensional dataset containing more than 8,000 valid samples.

[0027] S3. Model Training and Construction: Based on the above dataset, a multi-module fusion machine learning R&D model is trained, and a material property prediction proxy model is enabled to fit the mapping relationship between the proportion of five-element metal components, the concentration of microscopic defects, and the catalytic activity and structural stability of OER. The VAE-GAN module is equipped to extract the core structural features of high-entropy oxides through VAE dimensionality reduction, and GAN intelligently generates 20 groups of novel five-element high-entropy oxide candidate structures. Machine learning atomic force field and deep learning representation modules are configured to realize atomic mechanism analysis and intelligent analysis of electron microscopy images. An adaptive active learning iterative framework is enabled, and an experimental feedback iterative interface is reserved.

[0028] S4. Input target performance indicators: Set the target indicators for oxygen evolution catalysis in water electrolysis: overpotential ≤280 mV at a current density of 10 mA / cm², Tafel slope ≤45 mV / dec, and stability decay ≤3% after 1000 h of cycling.

[0029] S5. Model Simulation Result Verification: Model simulation results show that Ni2Co1Fe1Mn1Cu1O is obtained through screening. x High-entropy oxide is the optimal candidate system. Through atomic force field simulation, it is verified that the lattice distortion of this system is moderate, the number of surface hydroxyl active sites is abundant, it is compatible with the OER reaction mechanism, and the simulation performance indicators meet the preset requirements, so it is determined that it can be put into trial production.

[0030] S6. Initial Sample Preparation: Referring to the model output parameters, the target high-entropy oxide was prepared using the sol-gel method: A precursor solution with a total metal ion concentration of 0.5 mol / L was prepared, and the mixture was mixed according to the molar ratio of Ni:Co:Fe:Mn:Cu=2:1:1:1:1. After stirring evenly, the mixture was sol-gelled, dried at 120℃ for 12 h, sintered at 550℃ for 3 h, and naturally cooled to obtain the initial high-entropy oxide sample.

[0031] S7. Performance Testing and Parameter Optimization: The initial sample was tested, and the measured overpotential at 10 mA / cm² was 296 mV, and the Tafel slope was 48 mV / dec, indicating that the performance was not optimal. SEM and TEM deep learning analysis revealed uneven local component distribution and low microporosity in the sample. Based on model inverse optimization, the metal composition ratio was adjusted to Ni1.8Co1.2Fe1Mn1Cu1O. x The sintering temperature was increased to 580℃, and the holding time was extended to 3.5 h to optimize the micropore structure and composition uniformity.

[0032] S8. Secondary trial production: Samples are prepared again using optimized proportions and process parameters.

[0033] S9. Iterative optimization: Repeat the performance test and process optimization steps. After three iterations, the high-entropy oxide prepared finally has an overpotential of 272 mV at a current density of 10 mA / cm², a Tafel slope of 42 mV / dec, and a cycle stability decay of 1.8% after 1000 h. The performance fully meets the preset target and reaches the optimal state.

[0034] S10. Data backfilling and updating: Enter the optimal component ratio, sintering process parameters, microstructure parameters, and corresponding catalytic performance data into the database to complete the iterative update of the dataset.

[0035] Example 2: In-situ synthesis of high-entropy oxide electrocatalysts adapted for carbon dioxide electroreduction reaction (CO2RR) S1. Based on the existing database framework, expand the data storage module for carbon dioxide reduction catalysis.

[0036] S2. Data Expansion: By analyzing literature and patents in the field of CO2 electroreduction high-entropy catalytic materials using NLP technology, data on the preparation process, microstructure, catalytic selectivity, and Faraday efficiency of Cu-Zn-Sn-In-Ag pentagonal high-entropy oxides are supplemented; combined with atomic force field simulation, the adsorption and activation mechanism of CO2 by materials with different component ratios is analyzed, atomic-scale simulation data are supplemented, and the dataset is expanded to 7000+ samples.

[0037] S3. Model Training: Reuse multi-module machine learning R&D models, specifically train the CO2RR performance prediction branch, establish the mapping relationship between components, processes and CO product selectivity and Faraday efficiency, and generate a new high-entropy structure adapted to CO2 reduction through VAE-GAN.

[0038] S4. Input target indicators: Set CO2 electroreduction target: CO product Faradaic efficiency ≥92%, catalytic selectivity ≥90%, and 800 h cycle stability decay ≤2.5%.

[0039] S5. Simulation Verification: Model selection yielded Cu3Zn1Sn1In1Ag0.5O Atomic simulations show that the high-entropy oxide system has a moderate adsorption energy for CO2 intermediates on its surface and no side reaction sites, thus meeting the target performance requirements.

[0040] S6. Initial preparation: The sample was prepared by hydrothermal method with a total precursor concentration of 0.4 mol / L. The solution was prepared according to the initial model ratio, and the hydrothermal reaction was carried out at 150℃ for 8 h, followed by annealing at 450℃ for 2 h to obtain the initial sample.

[0041] S7-S9, Iterative Optimization: The initial sample CO Faradaic efficiency was 87.3%, which did not meet the preset target. Microscopic characterization revealed too many surface defects and numerous side reaction sites. Through adaptive iterative optimization using the model, the Ag doping ratio was adjusted to 0.8, the annealing temperature was increased to 480℃, and the surface defect structure was optimized. After two iterations, the finished sample achieved a CO Faradaic efficiency of 93.5%, a selectivity of 91.2%, and a stability decay of 2.1% after 800 h, demonstrating optimal performance.

[0042] S10. Fill the database with the optimal ratio, process parameters and performance data to complete the data iteration.

[0043] Example 3: In-house synthesis of high-entropy oxide electrocatalysts adapted for oxygen reduction reaction (ORR) in fuel cells This embodiment focuses on the oxygen reduction catalysis scenario of proton exchange membrane fuel cells, and utilizes a Co-Mn-Ni-Cr-V pentagonal high-entropy oxide system for independent synthesis. Data in the field of fuel cell catalysis is mined using NLP technology, and a specialized dataset is constructed using atomic simulations. After model training, the target indicators are: half-wave potential ≥ 0.85 V, limiting current density ≥ 5.2 mA / cm², and performance degradation ≤ 2% after 1000 cycles.

[0044] Model simulation was used to screen the optimal candidate structure. The structure was initially produced by solid-state sintering. The composition ratio and sintering process were iteratively optimized by combining electrochemical test and microscopic characterization results. The optimized high-entropy oxide had a half-wave potential of 0.88 V, a limiting current density of 5.5 mA / cm², and excellent cycle stability, which fully met the catalytic performance requirements for commercial fuel cells. Finally, the optimal process and ratio data were backfilled into the database.

[0045] Comparative test description The above three types of high-entropy oxide catalytic materials were developed using both traditional manual trial-and-error methods and the method of this invention. Traditional methods had a single-system development cycle of 30-45 days, and it was still difficult to reach the optimal state after 5-8 performance optimization iterations. In contrast, the method of this invention had a single-system development cycle of only 8-12 days, with ≤3 iterations, and the catalytic activity, stability, and selectivity were all 10%-20% better than those prepared by traditional processes, fully verifying the high efficiency and superiority of the method of this invention.

[0046] The above descriptions are merely some embodiments of the present invention. It should be noted that those skilled in the art can make other modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A method for the autonomous synthesis of high-entropy oxide electrocatalytic materials, characterized in that, Includes the following steps: S1. Construct a database of high-entropy oxide electrocatalytic materials; S2. Add multi-dimensional sample data to the high-entropy oxide electrocatalytic materials database, including microstructure parameters, preparation process parameters, electrocatalytic performance parameters, atomic simulation data, and scientific research text knowledge data. S3. Train and construct a multi-module integrated machine learning R&D model based on the electrocatalytic material database to establish the mapping relationship between material characteristics, process parameters and electrocatalytic performance; S4. Receive the preset target electrocatalytic performance indicators; S5. Verify the simulation analysis results of the analysis model; S6. Prepare high-entropy oxides by referring to the analysis results in step S5; S7. Analyze the properties of the finished high-entropy oxide and optimize the material ratio and preparation process parameters; S8. Prepare high-entropy oxides by referring to the analysis results in step S7; S9. Repeat steps S7 and S8 until the high-entropy oxide performs optimally. S10. Add the optimized material ratios and preparation process parameters to the high-entropy oxide electrocatalytic materials database.

2. The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to claim 1, characterized in that, The machine learning R&D model constructed in step S3 includes a material property prediction proxy model. The material property prediction proxy model is trained based on high-entropy oxide sample data in the dataset and fits the correlation mapping relationship between material composition, microstructure and electrocatalytic activity, structural stability and catalytic selectivity.

3. The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to claim 2, characterized in that, The machine learning development model is equipped with an adaptive active learning iterative framework. Based on the experimental feedback data of candidate materials, the active learning iterative framework continuously updates the model weight parameters, adaptively optimizes the high-entropy oxide composition ratio and preparation process parameters, and dynamically improves the model prediction accuracy and the performance matching degree of candidate materials.

4. The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to claim 3, characterized in that, The machine learning research model integrates a mutation autoencoder (VAE) and a generative adversarial network (GAN). The VAE is used to reduce the dimensionality, extract, and standardize the microstructural features of high-entropy oxides. The GAN intelligently generates multi-component novel high-entropy oxide candidate structures based on the characterization features, realizing innovative structural design and system expansion of high-entropy electrocatalytic materials.

5. The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to claim 4, characterized in that, The machine learning research model is configured with a machine learning atomic force field, and atomic-scale simulation calculations of high-entropy oxides are carried out through the machine learning atomic force field. Based on the simulation results, the evolution laws of lattice distortion, component diffusion, and surface active sites of high-entropy oxides are analyzed, and the intrinsic correlation mechanism between the configuration entropy of high-entropy oxides and the electrocatalytic reaction mechanism is established.

6. The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to claim 5, characterized in that, The machine learning R&D model integrates a deep learning representation module, which is used to collect and analyze atomic-scale microscopic imaging data of high-entropy oxides, intelligently identify and accurately analyze transmission electron microscopy and scanning electron microscopy imaging data, and quantitatively characterize the microscopic morphology, defect structure and multi-component distribution characteristics of the material.

7. The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to claim 6, characterized in that, Natural language processing (NLP) technology is integrated into the construction of the dataset. Through NLP, scientific research literature, patent texts, and experimental reports in the field of high-entropy materials are structured and parsed to automatically extract the related knowledge of material composition, preparation process, and performance characteristics, thereby supplementing and improving the sample dimensions of the dataset and providing data support for machine learning model training.

8. The method for autonomous synthesis of high-entropy oxide electrocatalytic materials according to any one of claims 1-7, characterized in that, The target electrocatalytic performance indicators include any one or more of the following: water electrolysis catalytic performance indicators, fuel cell catalytic performance indicators, and carbon dioxide reduction catalytic performance indicators. The machine learning development model differentiates and selects suitable high-entropy oxide candidate material systems based on the indicator thresholds for different electrocatalytic application scenarios.