Method and system for evaluating performance of catalyst microstructure correlation

Abstract

The present application relates to the technical field of catalyst performance evaluation, and particularly relates to a catalyst microstructure related performance evaluation method and system. The method collects multi-scale microstructure characterization data and catalytic performance data of a catalyst sample, performs spatial domain and frequency domain collaborative analysis to determine a structure descriptor set. A quantitative correlation between the microstructure and macroscopic performance is determined based on the set and the performance data, and a correlation weight coefficient is assigned through sensitivity analysis, and a target microstructure feature is determined accordingly. The preparation process parameters are optimized with the feature as a constraint, a new sample is prepared, and the quantitative correlation is dynamically updated, and the present application realizes accurate evaluation of catalyst performance and directional optimization of the preparation process.

Description

Technical Field

[0001] This invention relates to the field of catalyst performance evaluation technology, and in particular to a method and system for evaluating the performance of catalysts based on their microstructure. Background Technology

[0002] In the field of catalyst development, establishing the correlation between microstructure and macroscopic properties is the core of guiding catalyst design and optimization. Current conventional practices typically rely on characterizing the catalyst's microstructure at a single or limited scale, such as observing morphology through electron microscopy, analyzing crystal structure through X-ray diffraction, or determining specific surface area and pore size distribution through nitrogen adsorption. These characterization methods can obtain the catalyst's structural parameters at specific dimensions.

[0003] After obtaining the structural parameters, conventional techniques attempt to correlate them with catalytic performance test data (such as conversion, selectivity, and stability). A common approach is to use statistical regression methods to find empirical mathematical relationships between single or multiple structural parameters and performance indicators. This correlation analysis aims to provide clues for understanding the "structure-performance" relationship and hopefully guide adjustments to the preparation process.

[0004] However, the above-mentioned conventional approaches have significant drawbacks. The function of a catalyst stems from its complex microstructure across multiple scales and levels, and single or isolated characterization data cannot comprehensively and accurately describe the true distribution and dynamic evolution of active sites. For example, the nanoscale dispersion of active components, the mesoscale pore connectivity, and the atomic-scale electronic structure work synergistically to affect catalytic performance, but conventional methods lack effective means to synergistically analyze and integrate these multi-scale structural features.

[0005] Another drawback is that structure-performance correlation models built on limited data often lack robustness and predictive power. Because the weight of each microstructural feature's contribution to the final performance is not deeply quantified, the objective is not clear enough when using this correlation to guide process optimization. Adjustments to process parameters only improve minor structural features, while the key microstructural features that determine performance are not effectively controlled. This results in inefficient catalyst performance optimization processes with a degree of blindness and trial-and-error costs. Summary of the Invention

[0006] The present invention provides a method and system for evaluating the performance of catalysts based on their microstructure, which can solve the problems in the prior art.

[0007] A first aspect of the present invention provides a method for evaluating the performance of catalysts based on their microstructure, comprising:

[0008] Multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples were collected. Spatial and frequency domain co-analysis was performed on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution of active sites of the catalyst.

[0009] Based on the set of structural descriptors and the catalytic performance test data, the quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators is determined.

[0010] Based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation, a correlation weight coefficient characterizing the degree of influence of each structural descriptor on the catalytic performance is assigned, and the target microstructural features are determined based on the correlation weight coefficient.

[0011] Using the target microstructure characteristics as constraints, the catalyst preparation process parameters are iteratively adjusted to obtain an optimized process parameter scheme. A new catalyst sample is then prepared based on the optimized process parameter scheme, and the parameters in the quantitative correlation are dynamically updated based on the new catalyst sample.

[0012] Multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples were collected. Spatial and frequency domain co-analysis was performed on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution of active sites on the catalyst, including:

[0013] Multi-source characterization tests were performed on the catalyst sample to obtain morphological feature data, crystal structure feature data and electronic structure feature data as the multi-scale microstructure characterization data. The catalyst sample was then subjected to catalytic performance tests under preset reaction conditions to obtain catalytic performance test data.

[0014] The morphological feature data is spatially decomposed to determine the spatial frequency feature vector characterizing the spatial distribution uniformity of active sites, and the active site density distribution function is calculated based on the spatial frequency feature vector.

[0015] The electronic structure feature data are subjected to frequency domain transformation to determine the frequency domain response feature vector characterizing the electron binding energy state of active sites, and the electron affinity distribution function of active sites is calculated based on the frequency domain response feature vector.

[0016] The structure descriptor set is generated by coupling the active site density distribution function and the active site electron affinity distribution function.

[0017] Based on the set of structural descriptors and the catalytic performance test data, the quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators is determined, including:

[0018] Based on the spatial and energy descriptors in the set of structural descriptors, a high-dimensional feature tensor characterizing the coupling effect between the geometric configuration and electronic state of the active site is determined. The high-dimensional feature tensor is then projected in a dimensionality reduction direction to obtain a set of dominant feature vectors characterizing the spatial-energy synergistic effect of the active site.

[0019] The dominant feature vector set is subjected to temporal expansion to determine the dynamic feature trajectory characterizing the microstructure features as the catalytic reaction progresses;

[0020] Using the dynamic characteristic trajectory as the input variable and the catalytic performance test data as the output variable, a mapping function from the input variable to the output variable is determined through nonlinear regression calculation.

[0021] Based on the mapping function and the physical constraints based on the adsorption-desorption kinetic equilibrium conditions of the catalytic reaction, the dynamic characteristic trajectory of the catalyst sample to be evaluated is processed to obtain the predicted values ​​of catalytic activity and catalytic stability, so as to determine the quantitative correlation; wherein, the physical constraints include the fact that the direction of change of the predicted value of catalytic activity when the electron affinity potential of the active site changes is consistent with the direction of the effect of the change of the adsorption energy of the reactants.

[0022] Based on the spatial and energy descriptors in the structural descriptor set, a high-dimensional feature tensor characterizing the coupling effect between the geometric configuration and electronic state of active sites is determined. This high-dimensional feature tensor is then subjected to dimensionality reduction projection along a preset direction to obtain a set of dominant feature vectors characterizing the spatial-energy synergistic effect of active sites, including:

[0023] Multi-scale spatial decomposition is performed on the spatial descriptors in the structural descriptor set to determine the distribution characteristic components representing the active sites at different spatial scales;

[0024] The energy descriptors in the structure descriptor set are divided into energy level segments to determine the occupancy state characteristic components representing active sites in different electronic energy level intervals;

[0025] A high-dimensional feature tensor is determined by performing a tensor product operation based on the distribution feature components and the occupied state feature components.

[0026] Calculate the contribution variance of each coupling term in the high-dimensional feature tensor to the catalytic performance test data, and determine the dominant projection direction that retains the largest contribution variance based on the contribution variance;

[0027] The high-dimensional feature tensor is reduced in dimensionality and projected along the dominant projection direction. The preceding projection components whose cumulative contribution variance exceeds a preset threshold are extracted as the set of dominant feature vectors.

[0028] Based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation, a correlation weight coefficient characterizing the degree of influence of each structural descriptor on catalytic performance is assigned to each structural descriptor, and the target microstructural features are determined based on the correlation weight coefficient, including:

[0029] A preset perturbation amount is applied to each structure descriptor in the structure descriptor set to obtain the perturbed structure descriptor set.

[0030] Based on the perturbed set of structural descriptors and the quantitative correlation, the changes in the predicted catalytic activity and the predicted catalytic stability output by the quantitative correlation before and after the perturbation are calculated.

[0031] Based on the changes in the predicted catalytic activity and the predicted catalytic stability, the local and global sensitivity coefficients of each structural descriptor to the quantitative correlation output are calculated.

[0032] The comprehensive sensitivity index of each structural descriptor is obtained by weighted fusion of the local sensitivity coefficient and the global sensitivity coefficient. The association weight coefficient of each structural descriptor is obtained by processing the comprehensive sensitivity index.

[0033] The correlation weight coefficients are sorted according to their numerical values, and structural descriptors that meet preset conditions are selected as the target microstructural features based on the sorted correlation weight coefficients.

[0034] Based on the changes in the predicted catalytic activity and the predicted catalytic stability, the local and global sensitivity coefficients of each structural descriptor to the quantitative correlation output are calculated, including:

[0035] The proportional relationship between the contribution of catalytic activity and catalytic stability to the practical application performance of catalysts was determined based on the constraints of catalytic reaction kinetics.

[0036] The local sensitivity coefficient is obtained by coupling the change in the predicted catalytic activity value and the change in the predicted catalytic stability value based on the contribution ratio relationship.

[0037] For each structural descriptor, a multi-level sampling grid is constructed within the full numerical range of the structural descriptor. The preset perturbation amount is applied at the nodes of each level of the sampling grid to obtain the change in the predicted catalytic activity value and the change in the predicted catalytic stability value corresponding to each node.

[0038] Based on the changes in the predicted values ​​of catalytic activity and catalytic stability corresponding to each node, a nonlinear fitting is performed to determine the response surface function that characterizes the relationship between the numerical values ​​of the structure descriptor and the changes in the predicted values ​​of catalytic performance.

[0039] The global sensitivity coefficient is obtained by quantifying the non-monotonicity of the influence of the structure descriptor on catalytic performance based on the response surface function.

[0040] Using the target microstructure characteristics as constraints, the catalyst preparation process parameters are iteratively adjusted to obtain an optimized process parameter scheme. A new catalyst sample is then prepared based on this optimized scheme. The parameters in the quantitative correlation are dynamically updated based on the new catalyst sample, including:

[0041] The target microstructure features are transformed into quantitative constraints on the catalyst preparation process parameters, and the inverse mapping relationship between the catalyst preparation process parameters and the target microstructure features is determined.

[0042] Based on the inverse mapping relationship and the quantitative constraints, the catalyst preparation process parameters are iteratively solved, and the predicted value of the target microstructure features corresponding to the current preparation process parameters is calculated in each iteration.

[0043] Determine whether the predicted value of the target microstructure features meets the quantitative constraint conditions. If not, adjust the preparation process parameters according to the deviation direction between the predicted value of the target microstructure features and the quantitative constraint conditions, and proceed to the next iteration. If the conditions are met, determine the current preparation process parameters as the process parameter optimization scheme.

[0044] New catalyst samples were prepared according to the optimized process parameters, and multi-scale microstructure characterization data and catalytic performance test data were collected for the new catalyst samples.

[0045] The multi-scale microstructure characterization data and catalytic performance test data of the new catalyst sample are merged with historical catalyst sample data to obtain an extended sample dataset, and the parameters in the quantitative correlation are updated based on the extended sample dataset.

[0046] A second aspect of the present invention provides a catalyst microstructure-related performance evaluation system, comprising:

[0047] The data acquisition unit is used to acquire multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples, and to perform spatial and frequency domain joint analysis on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution state of active sites of the catalyst.

[0048] The correlation unit is used to determine the quantitative correlation between microstructure parameters and macroscopic catalytic performance indicators based on the set of structure descriptors and the catalytic performance test data.

[0049] The feature weighting unit is used to assign a correlation weight coefficient to each structural descriptor, characterizing its influence on catalytic performance, based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation relationship, and to determine the target microstructural features based on the correlation weight coefficient.

[0050] The process optimization unit is used to iteratively adjust the catalyst preparation process parameters using the target microstructure characteristics as constraints to obtain an optimized process parameter scheme, prepare a new catalyst sample according to the optimized process parameter scheme, and dynamically update the parameters in the quantitative correlation based on the new catalyst sample.

[0051] A third aspect of the present invention provides an electronic device, comprising:

[0052] processor;

[0053] Memory used to store processor-executable instructions;

[0054] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0056] This invention enables quantitative correlation analysis between the microstructure and macroscopic properties of catalysts. By co-analyzing multi-scale characterization data in the spatial and frequency domains, it can comprehensively capture key structural information such as the distribution, density, and spatial arrangement of active sites, forming a set of structural descriptors that accurately describe the microscopic state. These descriptors overcome the limitations of traditional single-parameter characterization, laying a data foundation for establishing reliable structure-activity relationship models.

[0057] This invention, based on the quantitative correlation established between structural descriptors and performance test data, can reveal the intrinsic mechanisms and contribution pathways by which microstructural parameters affect macroscopic catalytic performance. This model transforms complex catalytic processes into quantifiable mathematical relationships, enabling performance prediction and structural design. By performing sensitivity analysis on each structural descriptor and assigning correlation weights, the target microstructural features that play a dominant role in catalytic performance can be accurately identified.

[0058] This invention uses the target microstructural characteristics as constraints to guide the optimization of process parameters, achieving a closed-loop reverse design from performance requirements to preparation processes. This method can systematically and iteratively adjust preparation parameters, quickly obtaining process schemes that meet specific structural requirements, significantly shortening the catalyst development cycle. Preparing new samples based on the optimized scheme and dynamically updating model parameters ensures the timeliness and accuracy of the correlation, enabling the model to continuously improve itself.

[0059] Ultimately, this invention constructs a complete technology chain from microscopic characterization and correlation modeling to process optimization. It not only improves the accuracy and efficiency of catalyst performance prediction but also enables targeted control and rational design of the preparation process, providing strong methodological support for the development of high-performance catalysts. The entire process forms a data-driven research paradigm, promoting the digitalization and intelligentization of catalyst research and development. Attached Figure Description

[0060] Figure 1 This is a schematic flowchart of the catalyst microstructure correlation performance evaluation method according to an embodiment of the present invention;

[0061] Figure 2 This is a schematic diagram of the process for determining quantitative correlations according to an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only 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 scope of protection of the present invention.

[0063] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0064] Figure 1 This is a schematic flowchart of the catalyst microstructure correlation performance evaluation method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the catalyst microstructure-related performance evaluation methods include:

[0065] Multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples were collected. Spatial and frequency domain co-analysis was performed on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution of active sites of the catalyst.

[0066] Based on the set of structural descriptors and the catalytic performance test data, the quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators is determined.

[0067] Based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation, a correlation weight coefficient characterizing the degree of influence of each structural descriptor on the catalytic performance is assigned, and the target microstructural features are determined based on the correlation weight coefficient.

[0068] Using the target microstructure characteristics as constraints, the catalyst preparation process parameters are iteratively adjusted to obtain an optimized process parameter scheme. A new catalyst sample is then prepared based on the optimized process parameter scheme, and the parameters in the quantitative correlation are dynamically updated based on the new catalyst sample.

[0069] Multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples were collected. Spatial and frequency domain co-analysis was performed on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution of active sites on the catalyst, including:

[0070] Multi-source characterization tests were performed on the catalyst sample to obtain morphological feature data, crystal structure feature data and electronic structure feature data as the multi-scale microstructure characterization data. The catalyst sample was then subjected to catalytic performance tests under preset reaction conditions to obtain catalytic performance test data.

[0071] The morphological feature data is spatially decomposed to determine the spatial frequency feature vector characterizing the spatial distribution uniformity of active sites, and the active site density distribution function is calculated based on the spatial frequency feature vector.

[0072] The electronic structure feature data are subjected to frequency domain transformation to determine the frequency domain response feature vector characterizing the electron binding energy state of active sites, and the electron affinity distribution function of active sites is calculated based on the frequency domain response feature vector.

[0073] The structure descriptor set is generated by coupling the active site density distribution function and the active site electron affinity distribution function.

[0074] In practical applications, multi-source characterization tests were performed on supported metal catalyst samples using transmission electron microscopy (TEM), X-ray diffraction (XRD), and X-ray photoelectron spectroscopy (XPS). TEM was used to perform high-resolution imaging of the catalyst samples at an accelerating voltage of 200 kV, acquiring morphological data of the metal particles on the catalyst surface, including particle size distribution, interparticle spacing, and particle dispersion on the support surface. The image resolution was set to 0.1 nm, and the scanning area covered at least 50 fields of view to ensure statistical representativeness. XRD used a Cu Kα ray source, with a scanning angle range of 10° to 80°, a step size of 0.02°, and a scanning rate of 5° / min, to acquire crystal structure characteristic data of the catalyst, including diffraction peak positions, peak intensities, and full width at half maximum (FWHM). These data allowed for the determination of the metal phase composition, grain size, and degree of lattice distortion. X-ray photoelectron spectroscopy (XPS) was used to scan the surface elements of the catalyst under monochromatic Al Kα excitation. The binding energy scanning range was 0 to 1200 eV, and the resolution was set to 0.1 eV to acquire electronic structure characteristic data, including the chemical state of surface elements, binding energy position, and peak area ratio. Particular attention was paid to the valence state distribution information of metal active centers.

[0075] For catalytic performance testing, catalyst samples were loaded into a fixed-bed reactor with an inner diameter of 8 mm and a catalyst loading of 0.5 g. 1 cm of quartz sand was placed before and after the catalyst bed to ensure uniform gas flow distribution. The reaction temperature was set at multiple points within the range of 300℃ to 500℃, and data acquisition began after each temperature point had stabilized for 30 minutes. The reaction gas was a mixture of hydrogen and the target reactant, with a hydrogen flow rate of 30 mL / min and a reactant flow rate of 5 mL / min, precisely adjusted using a mass flow controller. The gas hourly space velocity (GHSV) was controlled at 3000 h⁻¹. The reaction products were analyzed in real-time using an online gas chromatograph equipped with a flame ionization detector (FID). Data was collected every 10 minutes for 8 consecutive hours to evaluate catalyst stability. Catalytic performance test data included reactant conversion, target product selectivity, and reaction rate per unit mass of catalyst at different reaction temperatures. Conversion was calculated based on the concentration difference of the reactant at the inlet and outlet, and selectivity was determined by the mole fraction of the target product among all products.

[0076] When performing spatial domain decomposition on morphological feature data, the two-dimensional image acquired by transmission electron microscopy is first converted into a digital matrix, with each pixel assigned a corresponding grayscale value. A two-dimensional Fourier transform is then used to decompose the spatial domain image into different frequency components, preserving amplitude and phase information during the transformation. The extraction of spatial frequency feature vectors is based on power spectral density analysis, calculating the frequency distribution characteristics in different directions. Low-frequency components correspond to large-scale morphological undulations on the catalyst surface, while high-frequency components reflect the microscopic distribution details of active sites. By radially integrating the power spectrum, characteristic parameters characterizing the spatial uniformity of active site distribution are obtained; higher parameter values ​​indicate more uniform dispersion of active sites on the support surface. The active site density distribution function is calculated by statistically analyzing the position coordinates of metal particles in the image and constructing a two-dimensional density field using a kernel density estimation method. A Gaussian kernel function is selected, and the bandwidth parameter is adaptively determined based on the average particle spacing. The density distribution function is expressed numerically as follows: , where x and y are two-dimensional coordinates of the carrier surface. The magnitude of this function directly reflects the enrichment degree of active sites at various locations in space. The total density of active sites can be obtained by integrating over the entire scanning area.

[0077] Frequency domain transformation of electronic structure characteristic data processes characteristic peaks in X-ray photoelectron spectroscopy, treating the binding energy spectrum as a one-dimensional signal with binding energy on the x-axis and photoelectron count intensity on the y-axis. A discrete Fourier transform is performed on this signal, transforming it to the frequency domain. The frequency domain response eigenvector contains the amplitude and phase of different frequency components. Low-frequency components correspond to the overall envelope of the spectral peaks, reflecting the distribution of the main chemical states, while high-frequency components correspond to the fine structure of the spectral peaks, revealing the existence of surface defect states and intermediate valence states. By analyzing the response intensity within a specific frequency range, the dispersion of the electronic binding energy states of active sites can be quantitatively characterized. The active site electron affinity distribution function is constructed based on the physical correlation between binding energy and electron affinity; peaks with lower binding energies correspond to active sites with higher electron affinity, making them more prone to adsorbing reactant molecules. The distribution function is expressed as... Where E is the binding energy, The value, after peak area normalization, represents the proportion of active sites in different binding energy states. The peak position and peak width of this function are directly related to the electronic properties of the catalyst.

[0078] The generation of the structure descriptor set achieves the fusion of spatial and electronic information through coupling operations, and incorporates the active site density distribution function. Distribution function of electron affinity Tensor product operations are performed to construct a three-dimensional descriptor space D(x,y,E). Each element in this space contains both spatial location and electronic state information; its numerical value represents the distribution density of active sites with specific electron affinities at a particular spatial location. To reduce the descriptor dimensionality, principal component analysis is performed on the three-dimensional space, and the top few principal components with a cumulative contribution rate reaching 95% are extracted as structural descriptors. Each principal component corresponds to a structural descriptor, whose physical meaning is a specific space-electronic state coupling mode. The set of structural descriptors is recorded in vector form. The descriptor denoted by n is the number of principal components, typically ranging from 5 to 15. These descriptors comprehensively reflect multi-dimensional information such as the spatial distribution uniformity, density, electronic state distribution width, and dominant electronic state type of active sites, providing a concise and complete input feature set for establishing quantitative correlations.

[0079] In practice, for catalyst samples from the same series prepared under different conditions, their structural descriptor sets and catalytic performance test data are acquired to establish a sample database. The database typically contains no fewer than 30 samples to ensure the reliability of statistical analysis. Each sample group includes a complete structural descriptor vector and corresponding catalytic performance indicators, including conversion, selectivity, and reaction rate constant at a specific reaction temperature. Through a systematic data acquisition and analysis process, an information bridge is built from microstructural characterization to macroscopic performance prediction, providing quantitative technical support for the rational design of catalysts.

[0080] Figure 2 This is a schematic diagram of the process for determining quantitative correlations according to an embodiment of the present invention, such as... Figure 2 As shown, based on the set of structure descriptors and the catalytic performance test data, the quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators is determined, including:

[0081] Based on the spatial and energy descriptors in the set of structural descriptors, a high-dimensional feature tensor characterizing the coupling effect between the geometric configuration and electronic state of the active site is determined. The high-dimensional feature tensor is then projected in a dimensionality reduction direction to obtain a set of dominant feature vectors characterizing the spatial-energy synergistic effect of the active site.

[0082] The dominant feature vector set is subjected to temporal expansion to determine the dynamic feature trajectory characterizing the microstructure features as the catalytic reaction progresses;

[0083] Using the dynamic characteristic trajectory as the input variable and the catalytic performance test data as the output variable, a mapping function from the input variable to the output variable is determined through nonlinear regression calculation.

[0084] Based on the mapping function and the physical constraints of the adsorption-desorption kinetic equilibrium conditions of the catalytic reaction, the dynamic characteristic trajectory of the catalyst sample to be evaluated is processed to obtain the predicted values ​​of catalytic activity and catalytic stability, so as to determine the quantitative correlation.

[0085] The physical constraint includes the fact that the direction of change of the predicted catalytic activity value of the mapping function when the electron affinity potential of the active site changes is consistent with the direction of the effect of the change in the adsorption energy of the reactants.

[0086] When constructing a quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators, it is necessary to fully consider the coupling effect between the geometric configuration and electronic states of active sites. Spatial descriptors obtained by scanning transmission electron microscopy can reflect the three-dimensional spatial distribution characteristics of active sites, including geometric information such as site spacing, coordination environment, and local symmetry. Meanwhile, energy descriptors obtained by X-ray photoelectron spectroscopy and synchrotron radiation absorption spectroscopy provide electronic structure information of active sites, including electron cloud density distribution, orbital hybridization degree, and density of states characteristics near the Fermi level. By fusing these two types of descriptors, a high-dimensional feature tensor containing multi-dimensional information such as spatial coordinates, electronic energy levels, and orbital occupancy numbers can be constructed.

[0087] The construction of this high-dimensional feature tensor involves the standardization of coordinate systems for data from different characterization techniques. The site coordinates in the spatial descriptor and the spectral peak information in the energy descriptor are mapped to the same reference frame, forming a third-order tensor of size M×7×P, where M represents the number of spatial sampling points, 7 represents the number of energy channels, and P represents the time or reaction condition dimension. This tensor contains complete state information of active sites at different reaction stages, but its high dimensionality significantly increases the computational complexity of subsequent modeling.

[0088] To reduce computational burden and extract key features, tensor decomposition is employed to perform dimensionality reduction projection on high-dimensional feature tensors. The direction of maximum variance is selected as the preset projection direction. By calculating the projection coefficients of the tensor along this direction, high-dimensional data is mapped to a low-dimensional feature space. Specifically, the covariance matrix of the tensor in each dimension is first calculated, and the first few principal component directions are determined through eigenvalue decomposition. The original tensor is then projected along these principal component directions, resulting in a set containing K dominant eigenvectors, each with a dimension of D. Here, K is much smaller than the total dimension of the original tensor, and D typically ranges from 10 to 50. These dominant eigenvectors comprehensively reflect the spatial-energy synergistic effect of active sites, preserving both the geometric arrangement information of the sites and the distribution characteristics of electronic states.

[0089] Catalytic reactions are dynamic evolution processes. The microstructure of active sites changes with reactant adsorption, surface reactions, and product desorption. To capture this dynamic evolution, the dominant feature vector set needs to be temporally expanded. During catalytic performance testing, multiple time points are set to collect catalyst microstructure data, with each time point corresponding to a set of dominant feature vectors. Arranging the feature vectors from different time points in chronological order forms a trajectory curve in the feature space; this trajectory curve is the dynamic feature trajectory. The morphological characteristics of the dynamic feature trajectory can reflect the evolution pattern of the active site state. For example, changes in trajectory curvature correspond to the reconstruction rate of the active site's electronic structure, while periodic oscillations reflect the cyclic characteristics of the adsorption-desorption process.

[0090] For long-term stability test data, the dynamic characteristic trajectory exhibits an initial rapid change followed by a gradual stabilization pattern. This pattern corresponds to the initial activation process on the catalyst surface and the subsequent steady-state operation stage. By analyzing the distribution region of the trajectory in the characteristic space, the stable operating state range of the catalyst can be identified. Significant deviation or divergence of the trajectory usually indicates that the catalyst is undergoing deactivation or structural collapse.

[0091] Establishing a quantitative relationship between dynamic characteristic trajectories and catalytic performance requires constructing a suitable mapping function. The eigenvectors at each time step of the dynamic characteristic trajectory are used as input variables, denoted as... The catalytic performance test data at the corresponding time point are used as output variables, including conversion rate. Selectivity Indicators such as... Since the relationship between microstructural features and macroscopic performance is usually highly nonlinear, a nonlinear regression method is used to construct a mapping function. Specifically, a kernel method can be used to map the input features to a high-dimensional space, establishing a linear relationship in this space. The mapping function can be expressed as... ,in For kernel function, For training sample points, Let L be the regression coefficient and L be the number of training samples. The regression coefficients are determined by minimizing the error between the predicted and actual test values, while a regularization term is introduced to prevent overfitting.

[0092] To ensure the physical rationality of the mapping function, physical constraints based on the catalytic reaction mechanism need to be introduced. Catalytic reactions follow the adsorption-desorption kinetic equilibrium principle, and the adsorption energy intensity of the active site on the reactants directly affects the catalytic activity. When the electron affinity of the active site increases, it means that the site is more likely to accept electrons, and the adsorption energy of electron-deficient reactant molecules will increase accordingly. According to the Sabatier principle, a moderate adsorption energy corresponds to optimal catalytic activity; excessively strong or weak adsorption is detrimental to the catalytic reaction. Therefore, the mapping function must satisfy the following constraint: when an increase in the characteristic parameter characterizing the electron affinity leads to an increase in adsorption energy, if the current site is in an under-adsorbed state, the predicted catalytic activity should increase; if the current site is in an over-adsorbed state, the predicted catalytic activity should decrease.

[0093] In practical calculations, physical constraints are implemented by adding a penalty term to the loss function of the mapping function. The partial derivatives of the mapping function with respect to the electron affinity-related characteristic parameters are calculated, and their signs are checked for consistency with theoretical expectations. When inconsistencies occur, a larger penalty weight is added to the sample point, forcing the model to adjust its parameters to meet the physical constraints. The mapping function corrected for physical constraints not only has better interpolation prediction capabilities but also higher reliability in extrapolating and predicting the performance of new catalysts.

[0094] For the catalyst sample to be evaluated, its multi-scale microstructure data were first obtained using the same characterization methods. Spatial and energy descriptors were extracted, a high-dimensional feature tensor was constructed, and dimensionality reduction projection was performed to obtain the dominant feature vector. The sample was monitored in real time under simulated catalytic reaction conditions, and the evolution trajectory of the feature vector over time was recorded, thus obtaining the dynamic feature trajectory of the sample. This dynamic feature trajectory was input into an established mapping function to obtain the predicted catalytic performance values ​​at each time point. Statistical analysis was performed on the predicted values, and the average performance value in the initial stage was taken as the predicted catalytic activity value. The predicted catalytic stability value was obtained by fitting the decay curve of the predicted values ​​over a long period. The predicted catalytic activity value reflects the intrinsic reactivity of the catalyst, while the predicted catalytic stability value characterizes the durability of the catalyst's performance maintenance.

[0095] Through the above processing steps, a quantitative correlation between microstructure descriptors and macroscopic catalytic performance indicators is finally established. This correlation not only includes the static structure-performance correspondence but also incorporates the influence of dynamic evolution processes, enabling a more comprehensive and accurate prediction of the actual performance of the catalyst. The establishment of this quantitative correlation provides a theoretical basis for subsequent catalyst design and optimization, allowing for precise control of target catalytic performance through the adjustment of microstructure parameters.

[0096] Based on the spatial and energy descriptors in the structural descriptor set, a high-dimensional feature tensor characterizing the coupling effect between the geometric configuration and electronic state of active sites is determined. This high-dimensional feature tensor is then subjected to dimensionality reduction projection along a preset direction to obtain a set of dominant feature vectors characterizing the spatial-energy synergistic effect of active sites, including:

[0097] Multi-scale spatial decomposition is performed on the spatial descriptors in the structural descriptor set to determine the distribution characteristic components representing the active sites at different spatial scales;

[0098] The energy descriptors in the structure descriptor set are divided into energy level segments to determine the occupancy state characteristic components representing active sites in different electronic energy level intervals;

[0099] A high-dimensional feature tensor is determined by performing a tensor product operation based on the distribution feature components and the occupied state feature components.

[0100] Calculate the contribution variance of each coupling term in the high-dimensional feature tensor to the catalytic performance test data, and determine the dominant projection direction that retains the largest contribution variance based on the contribution variance;

[0101] The high-dimensional feature tensor is reduced in dimensionality and projected along the dominant projection direction. The preceding projection components whose cumulative contribution variance exceeds a preset threshold are extracted as the set of dominant feature vectors.

[0102] When constructing a quantitative correlation between the microstructure and macroscopic properties of catalysts, there is a complex coupling effect between the geometric configuration and electronic state characteristics of active sites. In order to accurately characterize this coupling effect, it is necessary to process the spatial descriptors and energy descriptors in the set of structural descriptors in a coordinated manner to construct a high-dimensional feature tensor that can simultaneously reflect the spatial distribution characteristics and electronic energy states.

[0103] For the spatial descriptors in the structural descriptor set, a multi-scale spatial decomposition technique is employed. Specifically, the spatial domain of the catalyst sample is divided into three levels: atomic, nanoscale, and mesoscale. At the atomic scale, the spatial descriptor mainly includes the coordinate information of active site atoms, the nearest neighbor atomic spacing, and coordination geometry parameters. At the nanoscale, the spatial descriptor involves the morphological characteristics of active site clusters, surface roughness, and pore structure parameters. At the mesoscale, the spatial descriptor covers parameters such as the morphological characteristics of the support material and the macroscopic distribution uniformity of the active components. For each spatial scale, wavelet decomposition is used to extract the corresponding distribution feature components. Taking the nanoscale as an example, a two-dimensional discrete wavelet transform is performed on the transmission electron microscopy image, with a decomposition level of 5. In each decomposition level, four sub-band coefficients are obtained, corresponding to the low-frequency approximation information, horizontal high-frequency details, vertical high-frequency details, and diagonal high-frequency details of the image, respectively. By statistically analyzing the energy distribution of each sub-band coefficient, feature components characterizing the spatial non-uniformity of the active sites at the nanoscale are obtained. For the atomic scale, extended X-ray absorption fine structure spectral data are used, and Fourier transform is performed based on the radial distribution function to extract structural parameters of different coordination shells as distribution feature components. For the mesoscale scale, texture feature parameters such as contrast, correlation, energy, and uniformity are calculated through gray-level co-occurrence matrix analysis of scanning electron microscopy images as distribution feature components at this scale.

[0104] For energy descriptors in the structural descriptor set, energy level segmentation is performed. Based on X-ray photoelectron spectroscopy and ultraviolet photoelectron spectroscopy data, the electronic energy level range is divided into multiple intervals. Taking transition metal catalysts as an example, the valence band electronic energy level is divided into core energy level intervals, inner valence electron energy level intervals, and outer valence electron energy level intervals. The core energy level interval corresponds to the binding energy range of inner-shell electrons, and the peak position shift in this interval reflects the chemical valence state change of the active site. The inner valence electron energy level interval corresponds to the d-orbital electron distribution of the transition metal, and the peak intensity and peak shape characteristics in this interval reflect the d-electron occupancy state. The outer valence electron energy level interval is close to the Fermi level, and the density of states distribution in this interval directly affects the electron transfer process in the catalytic reaction. For each energy level interval, the integrated intensity, peak position, half-maximum width, and peak shape asymmetry factor of the electronic density of states within that interval are calculated as occupancy state characteristic components. In addition, for surface adsorbed species, desorption peaks in different temperature ranges were obtained through temperature-programmed desorption experiments. Each desorption peak corresponds to an adsorption state with a specific adsorption energy. The adsorption energy range was divided into intervals of 50 kJ / mol, and the number density of adsorption sites in each energy range was statistically analyzed as a supplementary occupancy state characteristic component.

[0105] After obtaining the distribution feature components at each spatial scale and the occupied state feature components at each energy level interval, tensor product operations are performed to construct a high-dimensional feature tensor. Let the distribution feature component vector at the i-th spatial scale be... Its dimensions are The occupancy state feature component vector of the j-th energy level interval is Its dimensions are Tensor product operations are performed on all combinations of spatial scales and energy level intervals to generate a fourth-order tensor. Its elements are defined as the product of different spatial scale components and energy level components. The dimension of this tensor is... When considering three spatial scales and three energy level intervals, if the eigencomponents of each scale and energy level have dimensions of 8 and 6 respectively, the total dimension of the complete tensor can reach several thousand dimensions. Each element in the high-dimensional eigentensor represents the coupling strength between the geometric distribution characteristics at a specific spatial scale and the electron occupancy state in a specific energy level interval. This coupling strength is directly related to the space-energy synergistic effect of the active sites of the catalyst during the reaction process.

[0106] To extract the dominant features influencing catalytic performance from the high-dimensional feature tensor, it is necessary to calculate the contribution variance of each coupling term in the tensor to the catalytic performance test data. The high-dimensional feature tensor is expanded into a matrix form, with row indices corresponding to sample numbers and column indices corresponding to different coupling terms. For a dataset containing 50 catalyst samples, the feature matrix has a size of 50 rows multiplied by thousands of columns. The covariance between each column feature and the catalytic performance index is calculated; the absolute value of the covariance reflects the explanatory power of that coupling term for performance variations. The sum of squared covariances of all coupling terms is normalized to obtain the contribution variance percentage of each coupling term. Analysis of variance reveals that over 70% of the performance variance can be explained by less than 15% of the coupling terms, indicating the existence of a dominant space-energy coupling mode.

[0107] Based on the contribution variance of each coupling term, the dominant projection direction that retains the largest contribution variance is determined. Principal component analysis (PCA) is used to perform singular value decomposition on the eigenma matrix to obtain orthogonal projection directions. The first dominant projection direction corresponds to the largest singular value; the projection component in this direction captures the largest variance information in the dataset. Subsequent projection directions correspond to decreasing singular values, with each direction capturing the largest portion of the remaining variance. In practice, the first 20 dominant projection directions are calculated, and the proportion of variance explained by each direction is statistically analyzed. The results show that the cumulative explained variance of the first 5 dominant projection directions reaches 85%, and the cumulative explained variance of the first 10 directions exceeds 95%.

[0108] The high-dimensional feature tensor is dimensionality-reduced by projecting along a defined dominant projection direction. The original feature matrix is ​​multiplied by the unit vectors of each dominant projection direction to obtain the dimensionality-reduced projection components. A preset threshold of 90% is set, selecting the preceding projection components whose cumulative explained variance reaches 90%. Based on the aforementioned statistical results, retaining the first 7 projection components satisfies this threshold requirement. These 7 projection components constitute the dominant feature vector set, and the representation of each sample in this set is given by a 7-dimensional vector, achieving significant dimensionality reduction compared to the original thousands of dimensions of features. Each component in the dominant feature vector set corresponds to a specific space-energy synergy mode. For example, the first dominant component mainly reflects the synergistic effect of active site cluster morphology and d-electron occupancy state at the nanoscale, while the second dominant component mainly reflects the synergistic effect of coordination geometry and surface adsorption energy distribution at the atomic scale.

[0109] Validated on multiple catalyst samples, the performance prediction model constructed using the dimensionality-reduced dominant feature vector set reduced the root mean square error of prediction on the test set by 40% compared to using all original features, while also shortening the model training time by 80%. This indicates that the extracted dominant feature vector set effectively retains the key information characterizing catalytic performance, removes redundant and noisy features, and provides high-quality feature input for subsequently establishing a precise quantitative correlation between microstructure and macroscopic performance.

[0110] Based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation, a correlation weight coefficient characterizing the degree of influence of each structural descriptor on catalytic performance is assigned to each structural descriptor, and the target microstructural features are determined based on the correlation weight coefficient, including:

[0111] A preset perturbation amount is applied to each structure descriptor in the structure descriptor set to obtain the perturbed structure descriptor set.

[0112] Based on the perturbed set of structural descriptors and the quantitative correlation, the changes in the predicted catalytic activity and the predicted catalytic stability output by the quantitative correlation before and after the perturbation are calculated.

[0113] Based on the changes in the predicted catalytic activity and the predicted catalytic stability, the local and global sensitivity coefficients of each structural descriptor to the quantitative correlation output are calculated.

[0114] The comprehensive sensitivity index of each structural descriptor is obtained by weighted fusion of the local sensitivity coefficient and the global sensitivity coefficient. The association weight coefficient of each structural descriptor is obtained by processing the comprehensive sensitivity index.

[0115] The correlation weight coefficients are sorted according to their numerical values, and structural descriptors that meet preset conditions are selected as the target microstructural features based on the sorted correlation weight coefficients.

[0116] After constructing the set of structural descriptors and establishing the quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators, it is necessary to further identify key microstructural features that have a significant impact on catalytic performance. This process is achieved through systematic sensitivity analysis, which can quantitatively reveal the differences in the contribution of structural descriptors of different scales and types to the overall performance of the catalyst.

[0117] For each independent descriptor in the defined set of structural descriptors, a perturbation of a specific magnitude is applied to simulate the structural fluctuations that occur in actual catalysts. The setting of the perturbation amount needs to comprehensively consider the physical meaning and numerical range of the structural descriptor. For descriptors characterizing pore size distribution, the perturbation amount can be set to a value within ±5% to ±15% of its average value; for descriptors characterizing the degree of lattice distortion, the perturbation amount can adopt a lattice constant shift of ±0.02 to ±0.08 Å; for descriptors characterizing the density of surface active sites, the perturbation amount can be applied at a relative change rate of ±8% to ±20%. Perturbation methods include three modes: unidirectional positive perturbation, unidirectional negative perturbation, and bidirectional symmetric perturbation. Among them, bidirectional symmetric perturbation can more comprehensively reflect the nonlinear impact of structural descriptor changes on performance prediction. In actual operation, for the i-th structural descriptor... The new value obtained after applying the perturbation is ,in This is the perturbation coefficient corresponding to the descriptor, and its value range is selected between -0.2 and 0.2 depending on the descriptor type.

[0118] The set of new structural descriptors obtained after perturbation is input into the established quantitative correlation model to calculate the predicted catalytic performance output under the perturbation state. The predicted catalytic performance output typically includes multiple dimensions of indicators. The predicted catalytic activity reflects the catalyst's conversion efficiency for the target reaction, which can be quantified by parameters such as conversion rate, reaction rate constant, or characteristic temperature. The predicted catalytic stability reflects the catalyst's ability to maintain performance during continuous use, which can be characterized by parameters such as activity decay rate, poisoning resistance index, or cycle life. For the i-th structural descriptor, the change in its predicted catalytic activity caused by perturbation is... Calculated as the predicted activity value after perturbation Compared with the initial activity prediction value The difference, that is Correspondingly, the change in the predicted catalytic stability value Calculated as ,in This is the predicted stability value after the disturbance. This represents the initial stability prediction value.

[0119] Based on the aforementioned changes in predicted values, the sensitivity coefficient of each structural descriptor to the quantitative correlation output is calculated. The local sensitivity coefficient reflects the degree of impact of small changes in the structural descriptor near the current operating point on performance prediction. The local sensitivity coefficient for the activity index is further specified. It can be represented as Local sensitivity coefficient for stability index It can be represented as This calculation method essentially involves numerical differentiation of the performance prediction function, capturing the immediate response characteristics of descriptor changes. The global sensitivity coefficient examines the cumulative impact of structural descriptors on performance prediction over a wider range of variations, obtained through multi-point sampling within a reasonable range of descriptor values. In practice, the value range of the i-th structural descriptor can be divided into several equally spaced or non-equally spaced sampling points. At each sampling point, the descriptor value is fixed while keeping other descriptors unchanged. The corresponding performance prediction output is calculated, and the variance contribution of the prediction outputs at all sampling points is statistically analyzed to quantify global sensitivity. Global sensitivity coefficient for activity metrics. It can be calculated using variance decomposition, and its value reflects the proportion of the contribution of the change of the i-th descriptor to the total variance of the activity prediction; the global sensitivity coefficient for the stability index. Obtained using the same principle.

[0120] After obtaining the local and global sensitivity coefficients, it is necessary to effectively fuse these two types of sensitivity information to comprehensively evaluate the importance of the structural descriptor. The fusion strategy adopts a weighted combination approach, which calculates the comprehensive sensitivity index of the activity dimension for the i-th structural descriptor. Calculated as ,in This is the fusion weighting coefficient for local and global sensitivity, typically ranging from 0.3 to 0.7. It can be adjusted based on the specific application requirements of the catalyst to determine either local precision or global robustness. For applications requiring precise performance control under specific operating conditions, Setting the value too high enhances the effect of local sensitivity; for applications that need to maintain performance reliability over a wide operating range, The value is set relatively low to highlight its contribution to global sensitivity. This is a comprehensive sensitivity index for the stability dimension. It was calculated using the same method.

[0121] Furthermore, it is necessary to integrate the comprehensive sensitivity indices of activity and stability dimensions into a single comprehensive sensitivity evaluation index. Considering the differences in the relative importance of activity and stability in practical applications of catalysts, the integration process also adopts a weighted summation method. The final comprehensive sensitivity index of the i-th structural descriptor is... Calculated as ,in This represents the relative importance weight between activity and stability, and its value is determined based on the specific application of the catalyst. For short-cycle, high-intensity catalytic processes, A value of 0.6 to 0.8 can be used to highlight the importance of activity; for catalytic processes operating continuously over long periods, A value of 0.3 to 0.5 can be used to emphasize the crucial role of stability.

[0122] The comprehensive sensitivity index is normalized and converted into correlation weight coefficients. Normalization methods can include linear normalization or exponential normalization. In linear normalization, the correlation weight coefficient of the i-th structural descriptor is... Calculated as Where M is the total number of descriptors in the structure descriptor set. Exponent normalization uses... In the form of, Temperature is a parameter used to adjust the concentration of the weight distribution; a smaller value indicates a higher concentration. The value will make the weight of highly sensitive descriptors more prominent.

[0123] All structural descriptors are sorted in descending order of their association weight coefficients to form an importance ranking sequence. Based on this ranking sequence, target microstructural features are selected using either a set threshold or a fixed number of parameters. The threshold selection method sets a minimum threshold for the association weight coefficients. All weight coefficients greater than or equal to The structural descriptor is identified as the target microstructural feature, and the threshold is used. The typical value range is 1.5 to 3 times the average weighting coefficient. A fixed-quantity screening method directly selects several top-ranked structural descriptors; the number can be set according to the complexity requirements of subsequent process optimization, generally selecting 5 to 15 descriptors to achieve a balance between feature representativeness and optimization operability. The selected set of target microstructural features will serve as the core constraint for subsequent catalyst preparation process parameter optimization, guiding the precise control of key structural elements during the preparation process.

[0124] Based on the changes in the predicted catalytic activity and the predicted catalytic stability, the local and global sensitivity coefficients of each structural descriptor to the quantitative correlation output are calculated, including:

[0125] The proportional relationship between the contribution of catalytic activity and catalytic stability to the practical application performance of catalysts was determined based on the constraints of catalytic reaction kinetics.

[0126] The local sensitivity coefficient is obtained by coupling the change in the predicted catalytic activity value and the change in the predicted catalytic stability value based on the contribution ratio relationship.

[0127] For each structural descriptor, a multi-level sampling grid is constructed within the full numerical range of the structural descriptor. The preset perturbation amount is applied at the nodes of each level of the sampling grid to obtain the change in the predicted catalytic activity value and the change in the predicted catalytic stability value corresponding to each node.

[0128] Based on the changes in the predicted values ​​of catalytic activity and catalytic stability corresponding to each node, a nonlinear fitting is performed to determine the response surface function that characterizes the relationship between the numerical values ​​of the structure descriptor and the changes in the predicted values ​​of catalytic performance.

[0129] The global sensitivity coefficient is obtained by quantifying the non-monotonicity of the influence of the structure descriptor on catalytic performance based on the response surface function.

[0130] In catalyst development, accurately assessing the influence of structural descriptors on catalytic performance is crucial for targeted regulation. To achieve this, it is necessary to quantify the impact of structural descriptors on quantitative correlation outputs from multiple dimensions. This process involves calculations at two levels: local perturbation response analysis and global parameter space mapping.

[0131] Regarding the predicted changes in catalytic activity and catalytic stability, it is first necessary to clarify their relative importance in practical catalytic applications. Catalytic activity determines the efficiency of catalyst conversion of reactants per unit time, while catalytic stability relates to the catalyst's ability to maintain performance over long-term operation. Based on catalytic reaction kinetic constraints, the contribution ratio of catalytic activity and catalytic stability to practical application performance is established. This constraint considers the influence of operating parameters such as reaction temperature, pressure, and space velocity on the catalyst deactivation rate. For high-temperature oxidation reaction systems, the catalyst is prone to thermal sintering, leading to the aggregation of active sites; in this case, the weight of catalytic stability should be increased. However, for selective hydrogenation reactions under mild conditions, the initial value of catalytic activity is often more critical. In specific calculations, by analyzing the deactivation kinetic equation of the catalyst under the target reaction conditions, the functional relationship between the activity decay rate constant and temperature and reactant concentration is determined, thereby deriving the proportion of catalytic activity loss on overall performance within the expected service life. Assuming that under a certain operating condition, the catalyst needs to be replaced when the conversion rate drops to 80% of the initial value, the cumulative conversion within this period can be calculated by integration to quantify the contribution coefficient of catalytic activity. Contribution coefficient to catalytic stability Both satisfy .

[0132] After obtaining the contribution ratio, the changes in predicted catalytic activity and predicted catalytic stability are coupled and calculated. Let the relative change in predicted catalytic activity after a perturbation of a certain structural descriptor be... The relative change in the predicted catalytic stability value is The overall performance response is then calculated as follows: Local sensitivity coefficient Defined as the ratio of the overall performance response to the applied perturbation. In practice, a small perturbation is applied to a structural descriptor near its current value point; the perturbation amplitude is typically set to 5% to 10% of the descriptor's standard deviation. For example, if a descriptor represents the distance between active sites, with a typical value of 8 nm and a standard deviation of 1.2 nm, the applied perturbation can be set to 0.06 to 0.12 nm. The predicted catalytic activity and catalytic stability values ​​output by the quantitative correlation model are calculated separately for positive and negative perturbations, and the changes are obtained using a differencing method. Substituting these two changes into the above coupling formula yields the overall performance response at the perturbation point. Local sensitivity coefficient. That is The quotient is calculated by dividing by the amount of perturbation. This coefficient reflects the immediate impact of a small change in the structure descriptor on overall catalytic performance near the current operating point.

[0133] However, local sensitivity coefficients can only reflect the linear response characteristics near specific numerical points and cannot capture the non-monotonic variations in the influence of structural descriptors on catalytic performance across the entire numerical domain. To obtain a global sensitivity assessment, systematic sampling analysis is needed across the complete range of structural descriptors. For each structural descriptor, the upper and lower bounds of its physically permissible numerical values ​​are first determined. For example, if a descriptor characterizes the fractal dimension of pore size distribution, its numerical range is theoretically defined between 1 and 3; if it characterizes the dispersion of metal nanoparticles, the numerical range is between 0 and 1. A multi-level sampling grid is constructed within the defined numerical range. The first level uses uniform intervals to divide the numerical range into 15 to 20 intervals, with the boundary points of each interval serving as sampling nodes. The second level adds sampling points at the midpoints of adjacent nodes in the first level, forming a denser grid. The third level further refines the grid resolution for the sensitivity abrupt change regions discovered in the first two levels of sampling, ensuring the capture of the transition characteristics of the nonlinear response.

[0134] At each node of the constructed multi-level sampling grid, a preset perturbation is applied to the structural descriptor. This perturbation operation is similar to that in local sensitivity analysis, but attention must be paid to the handling of boundary nodes: when a node is at the upper bound, only a negative perturbation is applied; when a node is at the lower bound, only a positive perturbation is applied; for internal nodes, both positive and negative perturbations are applied simultaneously. For each perturbation at each node, the changes in predicted catalytic activity and predicted catalytic stability are calculated using a quantitative correlation model. Since the sampling grid contains dozens or even hundreds of nodes, this process generates a large number of performance response data points. These data points form a discrete distribution in the coordinate space composed of the structural descriptor and the performance change, and their distribution pattern implies the global law of the descriptor's influence on catalytic performance.

[0135] Based on the predicted changes in catalytic activity and catalytic stability for each node, a coupled calculation is first performed according to the aforementioned contribution ratio, transforming the dual response values ​​of each node into a single comprehensive performance response index. Subsequently, a nonlinear fitting method is used to establish the functional relationship between the structural descriptor values ​​and the comprehensive performance response. The choice of fitting function needs to be determined based on the data distribution characteristics; commonly used forms include polynomial functions, Gaussian mixture functions, or radial basis functions. For response data exhibiting obvious unimodal or multimodal characteristics, Gaussian mixture functions can better describe their distribution patterns; the function form is a weighted sum of Gaussian components. For response data exhibiting periodic fluctuations, trigonometric function terms can be considered. The coefficient parameters in the fitting function are determined using the least squares method or other optimization algorithms to minimize the sum of squared residuals between the fitted curve and the discrete data points. After fitting, the fitting quality is evaluated, and the coefficient of determination is calculated to judge the interpretability of the fitting function for the actual data. This coefficient is required to be no less than 0.85 to ensure the reliability of the response surface function.

[0136] The obtained response surface function Describes the structure descriptor The evolution of the overall catalytic performance response across its full numerical domain is investigated. This function can be used to further quantify the non-monotonicity of the influence of the structure descriptor on catalytic performance. The first derivative of this function is calculated. and second derivative The non-monotonicity is assessed by analyzing the number of sign changes of the derivative. A change in the sign of the first derivative indicates the existence of an extreme point on the response surface, suggesting a reversal in the direction of the effect of increasing or decreasing the structural descriptor on catalytic performance. A change in the sign of the second derivative indicates a shift in the concavity / convexity of the curve, reflecting the changing characteristics of the performance response rate. A non-monotonicity index is defined. The absolute value of the curvature integral of the response surface function over the entire numerical domain is calculated using the following formula: ,in and These represent the lower and upper bounds of the structural descriptor, respectively. A larger integral value indicates a more pronounced curvature of the response surface, and a more complex nonlinear characteristic of the structural descriptor's influence on catalytic performance.

[0137] Global sensitivity coefficient Taking into account both the amplitude range and non-monotonicity of the response surface function, the calculation formula is as follows: ,in To respond to the range of the output value of the surface function across the entire numerical domain, and As a trade-off factor, it is usually determined based on the actual needs of catalyst development. For development scenarios that aim to maximize performance, A larger value is chosen to emphasize the magnitude of performance variation; for scenarios requiring precise control of performance stability, A larger value is chosen to focus on nonlinear response characteristics. The global sensitivity coefficient calculated in this way can comprehensively characterize the strength and complexity of the influence of the structure descriptor on catalytic performance throughout the entire parameter space, providing a quantitative basis for subsequent determination of correlation weight coefficients and identification of target microstructural features.

[0138] Using the target microstructure characteristics as constraints, the catalyst preparation process parameters are iteratively adjusted to obtain an optimized process parameter scheme. A new catalyst sample is then prepared based on this optimized scheme. The parameters in the quantitative correlation are dynamically updated based on the new catalyst sample, including:

[0139] The target microstructure features are transformed into quantitative constraints on the catalyst preparation process parameters, and the inverse mapping relationship between the catalyst preparation process parameters and the target microstructure features is determined.

[0140] Based on the inverse mapping relationship and the quantitative constraints, the catalyst preparation process parameters are iteratively solved, and the predicted value of the target microstructure features corresponding to the current preparation process parameters is calculated in each iteration.

[0141] Determine whether the predicted value of the target microstructure features meets the quantitative constraint conditions. If not, adjust the preparation process parameters according to the deviation direction between the predicted value of the target microstructure features and the quantitative constraint conditions, and proceed to the next iteration. If the conditions are met, determine the current preparation process parameters as the process parameter optimization scheme.

[0142] New catalyst samples were prepared according to the optimized process parameters, and multi-scale microstructure characterization data and catalytic performance test data were collected for the new catalyst samples.

[0143] The multi-scale microstructure characterization data and catalytic performance test data of the new catalyst sample are merged with historical catalyst sample data to obtain an extended sample dataset, and the parameters in the quantitative correlation are updated based on the extended sample dataset.

[0144] To transform the target microstructural features into quantitative constraints on catalyst preparation process parameters, it is necessary to establish a mathematical mapping relationship between microstructural features and process parameters. First, the target microstructural features are quantified, converting structural descriptors such as active site distribution density, pore size distribution characteristics, and lattice distortion degree into constraint parameters with defined numerical ranges. Taking supported noble metal catalysts as an example, if the target microstructural features require the average particle size of the noble metal particles to be controlled within the range of 2 to 5 nanometers, and the standard deviation of the particle size distribution to be less than 1.2 nanometers, then these values ​​are used as constraint boundary conditions. Simultaneously, it is necessary to identify the preparation process parameters directly related to these microstructural features, including but not limited to key parameters such as precursor solution concentration, impregnation temperature, calcination heating rate, holding time, and the hydrogen gas integral in the reducing atmosphere.

[0145] Establishing a reverse mapping relationship relies on a deep understanding of the catalyst formation mechanism. Taking the impregnation method for preparing supported catalysts as an example, the dispersion state of the precursor on the support surface is jointly affected by the surface tension of the solution, the capillary action of the support pores, and the solvent evaporation rate during the drying process. Microstructural data of catalyst samples under different combinations of process parameters are collected using experimental design methods to obtain the response surface between the preparation process parameters and the microstructural characteristics. Assuming the precursor solution concentration is C and the calcination temperature is... The heat preservation time is Target particle size is Then, the inverse mapping relationship can be established through empirical correlation using multiple regression analysis. When there are upper and lower limits for the target particle size, the constraints need to be expressed as the feasible region of process parameters, for example, requiring the particle size D to satisfy... At that time, the process parameter combination space that makes the particle size fall within the range can be solved by using the inverse mapping relationship.

[0146] The iterative solution process employs a numerical optimization algorithm to progressively adjust the process parameters. Initial process parameters can be selected from empirical values ​​or combinations of parameters corresponding to high-performing samples in historical data. In the first iteration, the initial process parameters are substituted into the forward prediction model to calculate the corresponding predicted values ​​of microstructural features. Based on established quantitative correlations, the forward prediction model outputs prediction results for microstructural parameters, including particle size distribution, specific surface area, and pore volume, after inputting the process parameters. The predicted microstructural features are compared with the target microstructural features, and the deviation of each structural descriptor is calculated. If the predicted particle size is... Target particle size is Then calculate the deviation. When the absolute value of the deviation exceeds the set tolerance threshold, it is determined that the current process parameters do not meet the constraints and parameter adjustments are required.

[0147] The adjustment strategy is determined based on the direction of deviation and the sensitivity coefficients of each process parameter to the microstructural characteristics. The sensitivity coefficients are obtained by taking the partial derivative of the inverse mapping relationship, representing the magnitude of microstructural characteristic change caused by a unit change in process parameter. Taking the effect of calcination temperature on particle size as an example, if sensitivity analysis shows that a 10°C increase in temperature leads to a 0.8 nm increase in particle size, the calcination temperature should be appropriately reduced when the predicted particle size is larger than the target particle size. The adjustment range is calculated based on the magnitude of the deviation and the sensitivity coefficients to avoid over-adjustment leading to oscillations. A damping factor is used to control the step size of each adjustment. A larger step size is used in the early stages of iteration to accelerate convergence, and the step size is reduced to improve solution accuracy as the solution approaches its optimum. For cases where multiple process parameters need to be adjusted simultaneously, the adjustment priority is allocated according to the relative magnitude of the sensitivity coefficients of each parameter, prioritizing the adjustment of parameters that have a significant impact on the target microstructural characteristics.

[0148] When determining whether the constraints are met, all target microstructural features need to be considered comprehensively. When multiple structural descriptor constraints exist, the predicted values ​​of all descriptors must fall within their respective target ranges simultaneously. If the particle size meets the constraint but the pore size distribution does not in a certain iteration, parameter adjustments continue. To prevent iterations from getting stuck in local optima or failing to converge, a maximum iteration limit is set. When the maximum number of iterations is reached but a combination of process parameters that satisfies all constraints is still not found, the tolerance range of some minor constraints is relaxed, or a multi-objective optimization strategy is introduced to weigh different structural features. After successfully finding process parameters that satisfy the constraints, the parameter combination is recorded as a process parameter optimization scheme, including a complete preparation process flow such as precursor solution concentration, impregnation method, drying conditions, calcination temperature curve, and reduction treatment parameters.

[0149] When preparing new catalyst samples according to the optimized process parameter scheme, each preparation step is strictly performed according to the parameters specified in the optimized scheme. Taking supported catalysts as an example, the solution is prepared according to the optimized precursor concentration, the impregnation temperature and time are controlled, the solvent is removed using an optimized drying procedure, and then calcination is completed according to the set heating rate and holding conditions. Key parameters are monitored in real time during the preparation process to ensure that the deviation between the actual process conditions and the optimized scheme is within the allowable range. After sample preparation, activation treatment is performed to bring the catalyst to a testable state.

[0150] Multi-scale microstructure characterization data collected from the new catalyst samples encompassed all characterization methods related to the target microstructure characteristics. The particle size distribution of noble metal particles was measured using transmission electron microscopy, and reliable particle size statistics were obtained from at least 200 particles. Specific surface area and pore size distribution were determined using nitrogen adsorption-desorption experiments to obtain pore volume data for mesopores and micropores. X-ray diffraction was used to analyze the crystal phase composition and grain size, and Raman spectroscopy was used to characterize the surface defect state. Catalytic performance test data were collected under standard reaction conditions, measuring key performance indicators such as conversion, selectivity, and specific activity, while simultaneously recording operating parameters such as reaction temperature, pressure, and space velocity.

[0151] When merging new catalyst sample data with historical sample data, it is necessary to ensure consistency in data format and testing conditions. Historical catalyst sample data includes the microstructural parameters and catalytic performance data of previously tested samples, forming the initial training dataset. New sample data is added to the dataset as incremental samples, expanding the coverage of the sample space, particularly increasing data density near the target microstructural features. The merged, expanded sample dataset contains more diverse structure-performance combinations, which helps improve the prediction accuracy and generalization ability of quantitative correlations.

[0152] When updating parameters in quantitative associations based on an expanded sample dataset, incremental learning is employed to avoid the computational burden of complete retraining. For quantitative associations established using machine learning models, the model parameters are fine-tuned using new sample data, and neural network weights or regression model coefficients are updated using optimization algorithms such as gradient descent. The contribution of historical samples to the model is retained during the update process to avoid catastrophic forgetting. For semi-empirical associations based on physical mechanisms, empirical coefficients are refitted using the expanded dataset, and weighted least squares is used to give new samples higher fitting weights, reflecting their guiding significance for the current optimization objective. After the parameter updates are completed, the updated model is validated using an independent validation set to assess whether the prediction error has decreased, confirming the improvement in model performance.

[0153] The dynamic update mechanism enables the quantitative correlation to continuously absorb new experimental data, gradually deepening the understanding of the catalyst structure-performance relationship. As optimization iterations proceed, the model's prediction accuracy for target microstructural feature regions continuously improves, providing more reliable guidance for subsequent catalyst design. Simultaneously, unexpected structural features or performance characteristics revealed by new sample data prompt a re-examination of key structural descriptors, driving the optimization and improvement of the structural descriptor set itself. The entire closed-loop iterative process achieves a complete cycle from microstructural analysis to performance prediction, then to process optimization and model updates, constructing a data-driven framework for rational catalyst design.

[0154] A second aspect of the present invention provides a catalyst microstructure-related performance evaluation system, comprising:

[0155] The data acquisition unit is used to acquire multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples, and to perform spatial and frequency domain joint analysis on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution state of active sites of the catalyst.

[0156] The correlation unit is used to determine the quantitative correlation between microstructure parameters and macroscopic catalytic performance indicators based on the set of structure descriptors and the catalytic performance test data.

[0157] The feature weighting unit is used to assign a correlation weight coefficient to each structural descriptor, characterizing its influence on catalytic performance, based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation relationship, and to determine the target microstructural features based on the correlation weight coefficient.

[0158] The process optimization unit is used to iteratively adjust the catalyst preparation process parameters using the target microstructure characteristics as constraints to obtain an optimized process parameter scheme, prepare a new catalyst sample according to the optimized process parameter scheme, and dynamically update the parameters in the quantitative correlation based on the new catalyst sample.

[0159] A third aspect of the present invention provides an electronic device, comprising:

[0160] processor;

[0161] Memory used to store processor-executable instructions;

[0162] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0163] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0164] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the performance of catalysts based on their microstructure, characterized in that, include: Multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples were collected. Spatial and frequency domain co-analysis was performed on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution of active sites of the catalyst. Based on the set of structural descriptors and the catalytic performance test data, the quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators is determined. Based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation, a correlation weight coefficient characterizing the degree of influence of each structural descriptor on the catalytic performance is assigned, and the target microstructural features are determined based on the correlation weight coefficient. Using the target microstructure characteristics as constraints, the catalyst preparation process parameters are iteratively adjusted to obtain an optimized process parameter scheme. A new catalyst sample is then prepared based on the optimized process parameter scheme, and the parameters in the quantitative correlation are dynamically updated based on the new catalyst sample.

2. The method according to claim 1, characterized in that, Multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples were collected. Spatial and frequency domain co-analysis was performed on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution of active sites on the catalyst, including: Multi-source characterization tests were performed on the catalyst sample to obtain morphological feature data, crystal structure feature data and electronic structure feature data as the multi-scale microstructure characterization data. The catalyst sample was then subjected to catalytic performance tests under preset reaction conditions to obtain catalytic performance test data. The morphological feature data is spatially decomposed to determine the spatial frequency feature vector characterizing the spatial distribution uniformity of active sites, and the active site density distribution function is calculated based on the spatial frequency feature vector. The electronic structure feature data are subjected to frequency domain transformation to determine the frequency domain response feature vector characterizing the electron binding energy state of active sites, and the electron affinity distribution function of active sites is calculated based on the frequency domain response feature vector. The structure descriptor set is generated by coupling the active site density distribution function and the active site electron affinity distribution function.

3. The method according to claim 1, characterized in that, Based on the set of structural descriptors and the catalytic performance test data, the quantitative correlation between microstructural parameters and macroscopic catalytic performance indicators is determined, including: Based on the spatial and energy descriptors in the set of structural descriptors, a high-dimensional feature tensor characterizing the coupling effect between the geometric configuration and electronic state of the active site is determined. The high-dimensional feature tensor is then projected in a dimensionality reduction direction to obtain a set of dominant feature vectors characterizing the spatial-energy synergistic effect of the active site. The dominant feature vector set is subjected to temporal expansion to determine the dynamic feature trajectory characterizing the microstructure features as the catalytic reaction progresses; Using the dynamic characteristic trajectory as the input variable and the catalytic performance test data as the output variable, a mapping function from the input variable to the output variable is determined through nonlinear regression calculation. Based on the mapping function and the physical constraints based on the adsorption-desorption kinetic equilibrium conditions of the catalytic reaction, the dynamic characteristic trajectory of the catalyst sample to be evaluated is processed to obtain the predicted values ​​of catalytic activity and catalytic stability, so as to determine the quantitative correlation; wherein, the physical constraints include the fact that the direction of change of the predicted value of catalytic activity when the electron affinity potential of the active site changes is consistent with the direction of the effect of the change of the adsorption energy of the reactants.

4. The method according to claim 3, characterized in that, Based on the spatial and energy descriptors in the structural descriptor set, a high-dimensional feature tensor characterizing the coupling effect between the geometric configuration and electronic state of active sites is determined. This high-dimensional feature tensor is then subjected to dimensionality reduction projection along a preset direction to obtain a set of dominant feature vectors characterizing the spatial-energy synergistic effect of active sites, including: Multi-scale spatial decomposition is performed on the spatial descriptors in the structural descriptor set to determine the distribution characteristic components representing the active sites at different spatial scales; The energy descriptors in the structure descriptor set are divided into energy level segments to determine the occupancy state characteristic components representing active sites in different electronic energy level intervals; A high-dimensional feature tensor is determined by performing a tensor product operation based on the distribution feature components and the occupied state feature components. Calculate the contribution variance of each coupling term in the high-dimensional feature tensor to the catalytic performance test data, and determine the dominant projection direction that retains the largest contribution variance based on the contribution variance; The high-dimensional feature tensor is reduced in dimensionality and projected along the dominant projection direction. The preceding projection components whose cumulative contribution variance exceeds a preset threshold are extracted as the set of dominant feature vectors.

5. The method according to claim 1, characterized in that, Based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation, a correlation weight coefficient characterizing the degree of influence of each structural descriptor on catalytic performance is assigned to each structural descriptor, and the target microstructural features are determined based on the correlation weight coefficient, including: A preset perturbation amount is applied to each structure descriptor in the structure descriptor set to obtain the perturbed structure descriptor set. Based on the perturbed set of structural descriptors and the quantitative correlation, the changes in the predicted catalytic activity and the predicted catalytic stability output by the quantitative correlation before and after the perturbation are calculated. Based on the changes in the predicted catalytic activity and the predicted catalytic stability, the local and global sensitivity coefficients of each structural descriptor to the quantitative correlation output are calculated. The comprehensive sensitivity index of each structural descriptor is obtained by weighted fusion of the local sensitivity coefficient and the global sensitivity coefficient. The association weight coefficient of each structural descriptor is obtained by processing the comprehensive sensitivity index. The correlation weight coefficients are sorted according to their numerical values, and structural descriptors that meet preset conditions are selected as the target microstructural features based on the sorted correlation weight coefficients.

6. The method according to claim 5, characterized in that, Based on the changes in the predicted catalytic activity and the predicted catalytic stability, the local and global sensitivity coefficients of each structural descriptor to the quantitative correlation output are calculated, including: The proportional relationship between the contribution of catalytic activity and catalytic stability to the practical application performance of catalysts was determined based on the constraints of catalytic reaction kinetics. The local sensitivity coefficient is obtained by coupling the change in the predicted catalytic activity value and the change in the predicted catalytic stability value based on the contribution ratio relationship. For each structural descriptor, a multi-level sampling grid is constructed within the full numerical range of the structural descriptor. The preset perturbation amount is applied at the nodes of each level of the sampling grid to obtain the change in the predicted catalytic activity value and the change in the predicted catalytic stability value corresponding to each node. Based on the changes in the predicted values ​​of catalytic activity and catalytic stability corresponding to each node, a nonlinear fitting is performed to determine the response surface function that characterizes the relationship between the numerical values ​​of the structure descriptor and the changes in the predicted values ​​of catalytic performance. The global sensitivity coefficient is obtained by quantifying the non-monotonicity of the influence of the structure descriptor on catalytic performance based on the response surface function.

7. The method according to claim 1, characterized in that, Using the target microstructure characteristics as constraints, the catalyst preparation process parameters are iteratively adjusted to obtain an optimized process parameter scheme. A new catalyst sample is then prepared based on this optimized scheme. The parameters in the quantitative correlation are dynamically updated based on the new catalyst sample, including: The target microstructure features are transformed into quantitative constraints on the catalyst preparation process parameters, and the inverse mapping relationship between the catalyst preparation process parameters and the target microstructure features is determined. Based on the inverse mapping relationship and the quantitative constraints, the catalyst preparation process parameters are iteratively solved, and the predicted value of the target microstructure features corresponding to the current preparation process parameters is calculated in each iteration. Determine whether the predicted value of the target microstructure features meets the quantitative constraint conditions. If not, adjust the preparation process parameters according to the deviation direction between the predicted value of the target microstructure features and the quantitative constraint conditions, and proceed to the next iteration. If the conditions are met, determine the current preparation process parameters as the process parameter optimization scheme. New catalyst samples were prepared according to the optimized process parameters, and multi-scale microstructure characterization data and catalytic performance test data were collected for the new catalyst samples. The multi-scale microstructure characterization data and catalytic performance test data of the new catalyst sample are merged with historical catalyst sample data to obtain an extended sample dataset, and the parameters in the quantitative correlation are updated based on the extended sample dataset.

8. A catalyst microstructure-related performance evaluation system, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to acquire multi-scale microstructure characterization data and corresponding catalytic performance test data of catalyst samples, and to perform spatial and frequency domain joint analysis on the multi-scale microstructure characterization data to determine the set of structural descriptors characterizing the distribution state of active sites of the catalyst. The correlation unit is used to determine the quantitative correlation between microstructure parameters and macroscopic catalytic performance indicators based on the set of structure descriptors and the catalytic performance test data. The feature weighting unit is used to assign a correlation weight coefficient to each structural descriptor, characterizing its influence on catalytic performance, based on the sensitivity analysis results of the contribution of each structural descriptor in the set of structural descriptors to the output of the quantitative correlation relationship, and to determine the target microstructural features based on the correlation weight coefficient. The process optimization unit is used to iteratively adjust the catalyst preparation process parameters using the target microstructure characteristics as constraints to obtain an optimized process parameter scheme, prepare a new catalyst sample according to the optimized process parameter scheme, and dynamically update the parameters in the quantitative correlation based on the new catalyst sample.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images