Microwave-assisted copper ferrite denitration catalyst performance prediction method based on data driving

Abstract

The invention discloses a microwave-assisted copper ferrite denitration catalyst performance prediction method based on data driving, and belongs to the field of neural networks, and the method comprises the steps: obtaining a microscopic image set of a copper ferrite denitration catalyst before microwave treatment; determining a local field intensity distribution feature vector based on the microwave field parameters, extracting an initial structure feature vector from the microscopic image set, and performing matching in a pre-constructed microwave structure evolution mapping library according to the microwave field parameters to obtain a corresponding initial structure evolution trend vector; adjusting the initial structure evolution trend vector by taking the local field intensity distribution feature vector as a conditional weight to obtain a target structure evolution trend vector, and fusing the target structure evolution trend vector with the initial structure feature vector to obtain a process structure feature tensor; and inputting the process structure feature tensor into a neural network prediction model to obtain a prediction performance result. According to the method, the accuracy of microwave-assisted performance prediction of the copper ferrite denitration catalyst can be improved.

Description

Technical Field

[0001] This invention relates to the field of neural networks, and more particularly to a data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalysts. Background Technology

[0002] Microwave-assisted treatment can significantly modulate the microstructure (such as particle size, porosity, and defect states) of copper ferrite denitration catalysts through selective heating and alteration of local electric fields, thereby optimizing their denitration activity and stability. However, the interaction between microwaves and catalysts is a dynamic coupling process involving electromagnetic fields, thermal effects, and material responses. Traditional methods struggle to accurately describe its intrinsic laws, making it difficult to measure the final performance of microwave treatment. This hinders the escape from inefficient trial-and-error R&D models and the achievement of targeted design and process optimization of catalyst performance.

[0003] Existing technologies typically employ a static correlation method based on results: first, a set of microwave parameters (such as power, frequency, and time) are fixed to treat the catalyst, and then only the final microstructure after treatment is characterized (e.g., scanning electron microscopy images), and performance prediction is based on these characterization results. The fundamental flaw of this approach is that it completely ignores the dynamic evolution path of the microstructure during microwave treatment, as well as the spatial non-uniformity of the interaction between the electromagnetic field and the material. Therefore, it cannot reliably guide the performance prediction of new parameters or new material systems. Summary of the Invention

[0004] This invention provides a data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalysts, which can improve the accuracy of predicting the performance of microwave-assisted copper ferrite denitration catalysts.

[0005] One embodiment of the present invention provides a data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalysts, comprising:

[0006] Obtain a set of microscopic images of the copper ferrite denitration catalyst before microwave treatment; Based on preset microwave field parameters, a local field strength distribution feature vector corresponding to the microstructure space of the copper ferrite denitration catalyst is determined. An initial structural feature vector of the copper ferrite denitration catalyst is extracted from the microscopic image set and matched with a pre-constructed microwave structure evolution mapping library according to the microwave field parameters to obtain a corresponding initial structural evolution trend vector. The initial structural evolution trend vector is adjusted with the local field strength distribution feature vector as a conditional weight to obtain a target structural evolution trend vector. The target structural evolution trend vector is then fused with the initial structural feature vector to obtain a process structural feature tensor reflecting the interaction between the microwave field and the catalyst structure. The microwave structure evolution mapping library is obtained by processing time-series structural image sequences under different microwave field parameters. The process structure feature tensor is input into a pre-trained performance prediction model to obtain the predicted performance results of the copper ferrite denitration catalyst after treatment under the microwave field parameters.

[0007] This invention establishes an accurate predicted material starting point by acquiring a set of microscopic images before microwave processing and extracting the initial structural feature vector, avoiding prediction bias caused by ignoring differences in the initial structure. Based on microwave field parameters, a local field strength distribution feature vector corresponding to the microscopic structure space is determined. This method accurately characterizes the non-uniform physical nature of the interaction between microwaves and catalysts, transforming macroscopic microwave parameters into real local driving conditions affecting structural evolution, fundamentally overcoming model errors caused by treating the microwave field as a uniform field. The initial structural evolution trend vector is obtained by matching the microwave field parameters from a pre-constructed microwave structural evolution mapping library. This method introduces universal dynamic laws of structural evolution under specific microwave parameters, revealed by historical experimental data, making prediction no longer limited to a single final static structure. Using the local field strength distribution feature vector reflecting spatial differences as conditional weights, the initial structural evolution trend vector is adjusted to obtain the target structural evolution trend vector. This innovative step achieves intelligent fusion of "universal dynamic laws" and "current sample-specific local conditions," enabling the prediction model to dynamically adapt to actual evolution paths under different microscopic environments. By fusing the target structure evolution trend vector with the initial structure feature vector, a process structure feature tensor is obtained. This tensor integrates the initial state, dynamic evolution law, and local field coupling effect, forming a high-dimensional information carrier that comprehensively represents the "microwave process-structure dynamic response." This process structure feature tensor is then input into a trained performance prediction model, enabling the model to make inferences based on the deeply fused features that reflect the entire process mechanism, thereby achieving more accurate and reliable predictions of the final performance.

[0008] Furthermore, the performance prediction model includes a spatiotemporal feature encoder, a multi-head attention fusion layer, and a regression prediction layer connected in sequence. The step of inputting the process structure feature tensor into the pre-trained performance prediction model to obtain the predicted performance results of the copper ferrite denitration catalyst after treatment under the microwave field parameters includes: The process structure feature tensor is input into a pre-trained performance prediction model, and the local structural features and spatiotemporal dependencies in the process structure feature tensor are extracted by the spatiotemporal feature encoder to construct the encoding result. The encoding result is processed by the multi-head attention fusion layer to obtain a dynamic evolution trend vector; The regression prediction layer performs a nonlinear mapping on the dynamic evolution trend vector to output the predicted performance results; wherein, the predicted performance results include denitrification efficiency and catalyst stability index.

[0009] By inputting the process structure feature tensor into the trained performance prediction model, the model can make inferences based on the deeply integrated features that reflect the entire process mechanism, thereby achieving a more accurate and reliable prediction of the final performance.

[0010] Further, determining the local field strength distribution characteristic vector corresponding to the microstructure space of the copper ferrite denitration catalyst based on preset microwave field parameters includes: The microwave field parameters are normalized to obtain normalized microwave parameters, wherein the microwave field parameters include microwave frequency, microwave power and microwave irradiation time. Based on the normalized microwave parameters, a reference field strength vector reflecting the average field strength level is calculated. Based on the porosity and composition distribution information reflected in the microscopic image set, the dielectric constant distribution feature vector at each location in the microstructure space is estimated through finite element electromagnetic simulation. The local field strength distribution feature vector is generated by performing element-wise multiplication between the reference field strength vector and the dielectric constant distribution feature vector.

[0011] This method, which determines the local field strength distribution characteristic vector corresponding to the microstructure space based on microwave field parameters, accurately characterizes the non-uniform physical nature of the interaction between microwaves and catalysts. It transforms macroscopic microwave parameters into real local driving conditions that affect structural evolution, fundamentally overcoming the model error caused by treating the microwave field as a uniform field.

[0012] Further, the step of performing element-wise multiplication of the reference field strength vector and the dielectric constant distribution feature vector to generate the local field strength distribution feature vector includes: Calculate the spatial autocorrelation coefficient of the dielectric constant distribution eigenvector; The dielectric constant distribution feature vector is weighted and smoothed based on the spatial autocorrelation coefficient to generate the correlation feature vector that reflects spatial dependence; The reference field strength vector and the associated feature vector are multiplied element-wise to generate the local field strength distribution feature vector.

[0013] Further, the extraction of the initial structural feature vector of the copper ferrite denitration catalyst from the microscopic image set includes: The microscopic image set is preprocessed to obtain a standardized microscopic image set, wherein the microscopic image set includes scanning electron microscope images and transmission electron microscope images; Based on image processing algorithms, multidimensional morphological features are extracted from the standardized microscopic image set to form the initial structural feature vector of the copper ferrite denitration catalyst; wherein, the morphological features include average particle size, size distribution variance, porosity and shape factor.

[0014] By standardizing the preprocessing of multi-source microscopic images and extracting multi-dimensional morphological features, a "digital twin" capable of comprehensively and quantitatively characterizing the initial microscopic state of the catalyst was constructed. This provides an accurate and consistent benchmark for subsequent evolutionary trend matching and adjustment, ensuring the reliability of the input basis for the entire prediction process.

[0015] Further, the step of matching the microwave field parameters with a pre-constructed microwave structure evolution mapping library to obtain the corresponding initial structure evolution trend vector includes: The microwave field parameters are matched with the index keys pre-stored in the microwave structure evolution mapping library to obtain the matching results; Based on the matching results, the associated temporal structure image sequence is determined; Calculate the difference vector sequence of structural features between adjacent time frames from the temporal structure image sequence; The difference vector sequence is fitted with a time series, and the feature parameters corresponding to the changing trend are extracted to form the initial structural evolution trend vector.

[0016] In this way, the initial structure evolution trend vector is obtained by matching the microwave field parameters from a pre-built microwave structure evolution mapping library. This method introduces the universal dynamic law of structure evolution under specific microwave parameters revealed by historical experimental data, so that the prediction is no longer limited to a single final static structure.

[0017] Furthermore, the microwave structure evolution mapping library is obtained by training time-series structure image sequences under different microwave field parameters, including: Using several combinations of initial microwave field parameters, the initial copper ferrite denitration catalyst samples were microwave-treated, and the initial time-series structure image sequence changing over time during the treatment process was acquired. The initial temporal structure image sequence is subjected to image denoising and feature alignment to obtain the processing result; Each of the initial microwave field parameters is combined as an index key, and the corresponding processing result is used as a data value to establish a one-to-one mapping relationship. Based on all the established mapping relationships, a microwave structure evolution mapping library is generated using the index key as the retrieval basis.

[0018] Further, the step of adjusting the initial structure evolution trend vector using the local field strength distribution feature vector as conditional weights to obtain the target structure evolution trend vector includes: Based on the local field strength distribution feature vector, calculate the weight distribution of the initial structure evolution trend vector; The initial structural evolution trend vector is weighted and fused according to the weight distribution to obtain the target structural evolution trend vector.

[0019] This achieves an intelligent fusion of "universal dynamic laws" and "current sample-specific local conditions," enabling the prediction model to dynamically adapt to the actual evolution path under different micro-environments.

[0020] Further, the process of fusing the target structure evolution trend vector with the initial structure feature vector to obtain the process structure feature tensor includes: Perform a tensor concatenation operation on the target structure evolution trend vector and the initial structure feature vector to obtain a concatenated feature tensor; The spliced ​​feature tensor is fused to output the process structure feature tensor.

[0021] By fusing the target structure evolution trend vector with the initial structure feature vector, a process structure feature tensor is obtained. This tensor integrates the initial state, dynamic evolution law, and local field coupling effect, forming a high-dimensional information carrier that comprehensively represents the "microwave process-structure dynamic response".

[0022] Furthermore, the training process of the performance prediction model includes: Obtain a training set, wherein the training set includes several samples, each sample including an initial process structure feature tensor and a corresponding true performance label; The training set is input into the initial performance prediction model. A multi-task joint learning strategy is adopted, and the parameters of the initial performance prediction model are optimized by the gradient descent algorithm to minimize the overall loss function. The performance prediction model is determined based on the optimized model parameters. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a schematic flowchart of an embodiment of the data-driven microwave-assisted copper ferrite denitration catalyst performance prediction method provided in this application; Figure 2 This is a flowchart illustrating one embodiment of steps S201 to S204 provided in this application; Figure 3 This is a flowchart illustrating one embodiment of steps S201 to S304 provided in this application; Figure 4 This is a flowchart illustrating one embodiment of steps S401 to S403 provided in this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0030] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0031] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0032] Microwave-assisted treatment can significantly modulate the microstructure (such as particle size, porosity, and defect states) of copper ferrite denitration catalysts through selective heating and alteration of local electric fields, thereby optimizing their denitration activity and stability. However, traditional methods struggle to accurately describe their intrinsic properties and lack effective methods for detecting the final performance of microwave treatment. Existing technologies typically employ result-based static correlation methods, but these cannot reliably guide performance prediction for new parameters or material systems.

[0033] See Figure 1 To improve the accuracy of performance prediction for microwave-assisted copper ferrite denitration catalysts, an embodiment of the present invention provides a data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalysts, comprising steps S101 to S103: Step S101: Obtain a set of microscopic images of the copper ferrite denitration catalyst before microwave treatment; In some embodiments, firstly, a powdered or shaped (e.g., granular, honeycomb) catalyst sample is prepared using any of the following processes: co-precipitation, sol-gel, or hydrothermal synthesis. Then, before microwave treatment begins, representative samples are randomly selected from the same batch of catalysts, and images are acquired using scanning electron microscopy (SEM) and transmission electron microscopy (TEM). SEM is used to acquire information on the microstructure, particle size distribution, and macroscopic pore structure of the catalyst surface and cross-section, while TEM is used to acquire higher-resolution information on grain morphology, lattice fringes, and internal microstructure. During acquisition, multiple fields of view are selected in different regions of the sample to ensure the statistical representativeness of the image set, and uniform imaging parameters (such as accelerating voltage, working distance, and magnification) are recorded. Finally, the obtained raw images are integrated to form a microscopic image set of the initial microstructure of the reaction catalyst.

[0034] It should be noted that for scanning electron microscopy (SEM) imaging, a small amount of powder sample is taken, evenly dispersed and fixed on conductive adhesive, and gold sputtering is performed if necessary to enhance conductivity; for transmission electron microscopy (TEM) imaging, the powder sample is ultrasonically dispersed in ethanol, the suspension is dropped onto a copper grid supported by an ultrathin carbon film and dried.

[0035] Step S102: Based on preset microwave field parameters, determine the local field strength distribution feature vector corresponding to the microstructure space of the copper ferrite denitration catalyst. Extract the initial structural feature vector of the copper ferrite denitration catalyst from the microscopic image set, and match it with the microwave field parameters in a pre-constructed microwave structure evolution mapping library to obtain the corresponding initial structural evolution trend vector. Adjust the initial structural evolution trend vector with the local field strength distribution feature vector as a conditional weight to obtain the target structural evolution trend vector. Then, fuse the target structural evolution trend vector with the initial structural feature vector to obtain a process structural feature tensor reflecting the interaction between the microwave field and the catalyst structure. The microwave structure evolution mapping library is obtained by processing time-series structural image sequences under different microwave field parameters. Please refer to Figure 2 In some embodiments, determining the local field strength distribution characteristic vector corresponding to the microstructure space of the copper ferrite denitration catalyst based on preset microwave field parameters includes steps S201 to S204: Step S201: Normalize the microwave field parameters to obtain normalized microwave parameters, wherein the microwave field parameters include microwave frequency, microwave power and microwave irradiation time. In some embodiments, the microwave field parameters are normalized to eliminate the influence of dimensions and scale to a uniform numerical range, resulting in normalized microwave parameters. Specifically, the microwave field parameters include the microwave frequency. (Unit: GHz), Microwave Power (Unit: W) and microwave irradiation time (Unit: s). Next, for each parameter, its maximum and minimum values ​​within the expected application range are determined. Subsequently, the collected raw parameter values ​​are linearly transformed using a minimum-maximum normalization method, mapping each parameter value to the interval [0, 1]. For example, for microwave power P, its normalized value... The normalized values ​​of the processed microwave frequency, microwave power, and microwave irradiation time together constitute the normalized microwave parameters. Other normalized microwave frequency values... Normalized value of microwave irradiation time It is also calculated using a similar formula, so I will not go into details.

[0036] Step S202: Based on the normalized microwave parameters, calculate the reference field strength vector that reflects the average field strength level. In some embodiments, firstly, the microwave frequency normalization value is... Normalized value of microwave power Normalized value of microwave irradiation time Through a preset fusion function This is mapped to a d-dimensional reference field strength vector. , that is The vector dimension d is consistent with the number of discretized grids in the microstructure space in subsequent steps.

[0037] It should be noted that the fusion function represents a nonlinear mapping relationship. It could be an empirical formula, such as a weighted summation containing nonlinear terms: In the formula, , , , , It is a learnable weight matrix or a constant matrix obtained from finite element simulation calibration, whose dimension d is consistent with the number of discretized grids in the microstructure space.

[0038] Step S203: Based on the porosity and composition distribution information reflected in the microscopic image set, the dielectric constant distribution feature vector at each location in the microstructure space is estimated through finite element electromagnetic simulation. In some embodiments, firstly, image analysis algorithms (such as thresholding and region growing) are used to quantitatively extract the porosity and distribution information of major elements such as iron, copper, and oxygen from the microscopic image set of the copper ferrite denitration catalyst. Then, based on this distribution information, a simplified three-dimensional geometric model reflecting the micropore and component distribution characteristics of the actual catalyst is constructed in finite element analysis software (such as COMSOL Multiphysics). In this model, based on the image analysis results, the intrinsic dielectric constant of copper ferrite (its real and imaginary parts can be obtained from material handbooks or experimental measurements) is assigned to the solid material region, and the dielectric constant of air (approximately equal to 1) is assigned to the pore region. Subsequently, the microwave frequency is set... The corresponding excitation port boundary conditions are then determined. Next, the electromagnetic simulation module of the software (such as the frequency domain solver) is run to obtain the numerical solutions of the complex relative permittivity at each discrete finite element mesh node within the simplified model. The solutions (real and imaginary parts) at all nodes are then arranged in mesh index order, thus forming the permittivity distribution characteristic vector. (Assume each node has two values: a real part and an imaginary part).

[0039] It should be noted that the intrinsic dielectric constant can be obtained from material handbooks or experimental measurements.

[0040] Step S204: Perform element-wise multiplication of the reference field strength vector and the dielectric constant distribution feature vector to generate the local field strength distribution feature vector.

[0041] In some embodiments, step S204 includes: calculating the spatial autocorrelation coefficient of the dielectric constant distribution feature vector; performing weighted smoothing processing on the dielectric constant distribution feature vector based on the spatial autocorrelation coefficient to generate the correlated feature vector reflecting spatial dependence; and multiplying the reference field strength vector and the correlated feature vector element-wise to generate the local field strength distribution feature vector. Specifically, firstly, based on the dielectric constant distribution feature vector... The spatial grid topology is reconstructed into a field map. Wherein, the dielectric constant value is given for each position i in the figure. (Can be real or imaginary part), calculate its relationship with its first-order neighborhood. Spatial autocorrelation coefficient of all positions j within the range This is used to quantify the degree of local statistical correlation of dielectric properties in the microscopic space. Then, using the calculated spatial autocorrelation coefficient as the weight basis, a spatial smoothing filter (e.g., a Gaussian kernel function whose weight distribution is adjusted by the autocorrelation coefficient) is constructed. This filter is then convolved on the spatial dimension of the dielectric constant distribution feature vector to utilize the spatial dependence of dielectric properties in adjacent regions, smoothing out potential noise or local extreme values ​​in the original simulation results, and enhancing the signal in spatially continuous regions. The final output is a correlation feature vector that preserves the macroscopic distribution pattern and explicitly implies the spatial dependence of the microscopic structure. Finally, the reference field strength vector Related feature vectors generated in this step The Hadamard product is performed so that the reference field strength, which reflects the external average energy level, is modulated by the dielectric property distribution inside the catalyst after spatial correlation correction, thereby generating a local field strength distribution feature vector that can more accurately characterize the non-uniform distribution of microwave field within the complex microstructure.

[0042] It should be noted that the spatial autocorrelation coefficient is used to quantify the statistical correlation of the dielectric constant within the microscopic neighborhood of the catalyst. Calculation method: In the formula, and Let i represent position i and its first-order neighborhood, respectively. The dielectric constant (real or imaginary part) at position j. This is the arithmetic mean of the dielectric constants at all locations within the entire analysis region. It is the variance of the dielectric constant over the entire region. It is a symmetric spatial weight matrix defined based on spatial adjacency relationships. If positions i and j are adjacent (e.g., sharing a mesh edge or vertex), then... ,otherwise .

[0043] It should be noted that the weighted smoothing process can be mathematically represented as a convolution operation, that is, associating feature vectors. It is possible = K ε is obtained, where, This represents a convolution operation, where K is a spatial smoothing filter kernel function, and its specific weight distribution is determined by the spatial autocorrelation coefficients at each location. Adjustments confirmed. This operation effectively incorporates the local spatial dependence of the dielectric constant.

[0044] It should be noted that the formula for calculating the Hadamard product is as follows: ,in, This indicates element-wise multiplication. yes The vector after appropriate selection and dimension adjustment This is the characteristic vector of the local field strength distribution.

[0045] This method, which determines the local field strength distribution characteristic vector corresponding to the microstructure space based on microwave field parameters, accurately characterizes the non-uniform physical nature of the interaction between microwaves and catalysts. It transforms macroscopic microwave parameters into real local driving conditions that affect structural evolution, fundamentally overcoming the model error caused by treating the microwave field as a uniform field.

[0046] In some embodiments, extracting the initial structural feature vector of the copper ferrite denitration catalyst from the microscopic image set includes: preprocessing the microscopic image set to obtain a standardized microscopic image set, wherein the microscopic image set includes scanning electron microscope images and transmission electron microscope images; and extracting multidimensional morphological features from the standardized microscopic image set based on an image processing algorithm to constitute the initial structural feature vector of the copper ferrite denitration catalyst; wherein the morphological features include average particle size, size distribution variance, porosity, and shape factor. Specifically, the microscopic image set is preprocessed to obtain a standardized microscopic image set. Subsequently, threshold segmentation (such as the Otsu method) or edge detection algorithms are used to segment catalyst particles and pore regions from the standardized microscopic image set. For each segmented independent particle region, its equivalent circle diameter is calculated as the particle size. The arithmetic mean of all particle sizes is calculated to obtain the average particle size, and its standard deviation or variance is calculated as the size distribution variance. The porosity is obtained by calculating the ratio of the pore region area to the total analyzed area of ​​the image. Simultaneously, the perimeter and area of ​​each particle region are calculated, and the formula (such as (perimeter²) / (4π)) is used to determine the porosity. The shape factor is calculated based on the area to characterize the degree to which the particle shape deviates from a circular shape. Finally, the four key indicators—average particle size, size distribution variance, porosity, and shape factor—are arranged and combined in a preset order to form a multi-dimensional numerical vector, which constitutes the initial structural feature vector of the copper ferrite denitrification catalyst.

[0047] It should be noted that the preprocessing process includes: converting all images to grayscale, using median filtering or Gaussian filtering algorithms to reduce image noise and eliminate random noise; normalizing the scale of all images, calibrating their pixel size to the same physical scale (e.g., nanometers / pixel) according to the scale information; and performing contrast-limited adaptive histogram equalization to enhance the visibility and consistency of structural features.

[0048] By standardizing the preprocessing of multi-source microscopic images and extracting multi-dimensional morphological features, a "digital twin" capable of comprehensively and quantitatively characterizing the initial microscopic state of the catalyst was constructed. This provides an accurate and consistent benchmark for subsequent evolutionary trend matching and adjustment, ensuring the reliability of the input basis for the entire prediction process.

[0049] In some embodiments, the microwave structure evolution mapping library is trained using time-series structure image sequences under different microwave field parameters, including: using several initial microwave field parameter combinations to microwave process initial copper ferrite denitration catalyst samples respectively, and collecting initial time-series structure image sequences that change over time during the processing; performing image denoising and feature alignment on the initial time-series structure image sequences to obtain processing results; establishing a one-to-one mapping relationship by using each initial microwave field parameter combination as an index key and the corresponding processing result as a data value; and generating the microwave structure evolution mapping library based on all established mapping relationships, using the index key as the retrieval basis. Specifically, firstly, a batch of copper ferrite denitration catalyst samples with the same initial state are prepared; then, several different initial microwave field parameter combinations are designed and used to process independent catalyst samples in a controllable microwave reaction device. During the processing, scanning electron microscopy and transmission electron microscopy are used to continuously image the same observation area of ​​the sample at preset, regular time intervals, thereby acquiring a set of initial time-series structure image sequences showing the continuous change of the microstructure of the reaction catalyst over time for each parameter combination. Subsequently, median filtering and other methods were used to denoise each initial time-series image sequence to improve the signal-to-noise ratio. Spatially, images from all time frames were aligned using feature point matching or image registration algorithms to ensure that images from different times observed the same microscopic region of the catalyst, resulting in a processed time-series image sequence with strictly aligned temporal and spatial features. Next, each initial microwave field parameter combination used for processing (e.g., in the form of a string "frequency_power_time") was defined as a unique index key, and its corresponding pre-processed and feature-aligned time-series structural image sequence was used as the associated data value, establishing a one-to-one key-value pair mapping relationship in a database or data structure. Finally, by summarizing and storing all such mapping relationships established through experiments with different parameter combinations, a structured microwave structure evolution mapping library was generated.

[0050] It should be noted that the microwave structure evolution mapping library uses microwave field parameter combinations as index keys, which can efficiently retrieve standard time-series image data of catalyst microstructure evolution under corresponding microwave treatment conditions.

[0051] It should be noted that each combination of initial microwave field parameters includes a specific microwave frequency, power, and irradiation time.

[0052] Please refer to Figure 3 In some embodiments, the step of matching the microwave field parameters with a pre-constructed microwave structure evolution mapping library to obtain the corresponding initial structure evolution trend vector includes steps S301 to S304: Step S301: Match the microwave field parameters with the index keys pre-stored in the microwave structure evolution mapping library to obtain the matching result; In some embodiments, the input microwave field parameters are formatted to form a string format identical to the index key of the mapping library (e.g., "F_P_T"). If a completely identical index key exists, the matching result is that key; if no complete match exists, the Euclidean distance between the input parameter vector and the parameter vectors corresponding to all index keys in the library is calculated, the index key with the smallest distance is selected as the best matching result, and the difference in matching parameters is recorded.

[0053] Step S302: Determine the associated temporal structure image sequence based on the matching result; In some embodiments, the microwave structure evolution map library is searched based on the matching results to obtain a pre-processed (denoised and aligned) temporal structure image sequence that is uniquely associated with the set of microwave field parameters.

[0054] Step S303: Calculate the difference vector sequence of structural features between adjacent time frames from the temporal structure image sequence; In some embodiments, firstly, for each frame of the temporal structured image sequence, a similar image processing algorithm is used to extract the same multidimensional morphological features (average particle size, size distribution variance, porosity, shape factor) to construct the structural feature vector for each frame. Then, in chronological order, the difference between the feature vector of the next frame and the feature vector of the previous frame is calculated, i.e. ,in, This represents the structural feature vector of frame t, and the difference between all adjacent time frames. Arranged in chronological order, they form a sequence of differential vectors describing the changes in structural features frame by frame.

[0055] Step S304: Perform time series fitting on the difference vector sequence, extract the feature parameters corresponding to the changing trend, and construct the initial structure evolution trend vector.

[0056] In some embodiments, for each feature dimension (such as particle size change) in the difference vector sequence, its change over time is treated as an independent time series. Trend modeling is then employed using models such as multinomial fitting or exponential decay fitting. For example, a first-order linear fit is used for particle size growth. The slope 'a' (average rate of change) is extracted as the key trend parameter. At the same time, this fitting process is repeated for all morphological feature dimensions. The extracted trend parameters of each dimension (such as the slope and intercept of the linear model) are then concatenated in sequence to form an initial structural evolution trend vector that comprehensively represents the overall direction and rate of change of structural features under given microwave parameters.

[0057] In this way, the initial structure evolution trend vector is obtained by matching the microwave field parameters from a pre-built microwave structure evolution mapping library. This method introduces the universal dynamic law of structure evolution under specific microwave parameters revealed by historical experimental data, so that the prediction is no longer limited to a single final static structure.

[0058] In some embodiments, adjusting the initial structure evolution trend vector using the local field strength distribution feature vector as conditional weights to obtain the target structure evolution trend vector includes: calculating the weight distribution of the initial structure evolution trend vector based on the local field strength distribution feature vector; and performing weighted fusion of the initial structure evolution trend vector according to the weight distribution to obtain the target structure evolution trend vector. Specifically, firstly, global average pooling and max pooling operations are performed on the local field strength distribution feature vector to extract its average field strength level and local hotspot intensity, respectively, and these two scalar values ​​are concatenated into a two-dimensional feature. Subsequently, this two-dimensional feature is input into a lightweight fully connected neural network layer, the output dimension of which is the same as the dimension of the initial structure evolution trend vector; simultaneously, the output is normalized using the Softmax function, i.e., generating a weight distribution vector whose sum of all elements is 1. , where this weight distribution vector The physical meaning lies in the fact that each of its components The importance of correcting corresponding dimensions (such as particle size change rate, porosity change rate, etc.) in the initial structural evolution trend vector based on the current microwave local field strength characteristics was quantified. Then, a weighted fusion was performed according to the weight distribution to obtain the target structural evolution trend vector. The obtained weight distribution vector was then... With the initial structural evolution trend vector Perform element-wise multiplication (Hadamard product), i.e. ,in Each element This operation enables targeted scaling adjustments to the general evolution trend (initial structural evolution trend vector) obtained based on macroscopic parameter matching, based on the actual energy distribution generated by the interaction between the microwave field and the catalyst's microstructure (characterized by local field strength characteristics). The final result... This is the target structure evolution trend vector, which more accurately predicts the evolution direction and intensity of the catalyst microstructure under the current specific microwave field, taking into account the spatial non-uniformity effect.

[0059] This achieves an intelligent fusion of "universal dynamic laws" and "current sample-specific local conditions," enabling the prediction model to dynamically adapt to the actual evolution path under different micro-environments.

[0060] In some embodiments, fusing the target structure evolution trend vector with the initial structure feature vector to obtain a process structure feature tensor includes: performing a tensor concatenation operation on the target structure evolution trend vector and the initial structure feature vector to obtain a concatenated feature tensor; and performing feature fusion on the concatenated feature tensor to output the process structure feature tensor. Specifically, firstly, the target structure evolution trend vector and the initial structure feature vector are used as inputs. Both vectors are one-dimensional numerical vectors, which can be connected along the last dimension (i.e., the feature dimension) in the program using tensor operation functions (such as torch.cat in PyTorch or tf.concat in TensorFlow) to merge them into a higher-dimensional concatenated feature tensor, where the tensor simultaneously encodes the catalyst's "static" initial structural endowment and "dynamic" predicted evolution trend. Subsequently, the feature tensor obtained by the above concatenation is input into a specially designed feature fusion module. This module typically consists of one or more fully connected layers, supplemented by a nonlinear activation function (such as ReLU). The core function of this fusion layer is to learn and integrate the complex interaction between the initial state and the evolution trend through nonlinear transformation, such as identifying the synergistic effect produced by the combination of a certain initial porosity and a specific grain growth trend. Finally, the fusion module outputs a new tensor with a fixed and uniform dimension, namely the process structure feature tensor.

[0061] It should be noted that the process structure feature tensor is a high-level characterization that is a deep integration of the initial structural features and the target evolution trend. It comprehensively reflects the final process structure state that may result from the interaction between specific microwave field parameters and the initial structure of a specific catalyst.

[0062] By fusing the target structure evolution trend vector with the initial structure feature vector, a process structure feature tensor is obtained. This tensor integrates the initial state, dynamic evolution law, and local field coupling effect, forming a high-dimensional information carrier that comprehensively represents the "microwave process-structure dynamic response".

[0063] Step S103: Input the process structure feature tensor into the pre-trained performance prediction model to obtain the predicted performance results of the copper ferrite denitration catalyst after treatment under the microwave field parameters.

[0064] In some embodiments, the training process of the performance prediction model includes: acquiring a training set, wherein the training set includes several samples, each sample including an initial process structure feature tensor and a corresponding true performance label; inputting the training set into the initial performance prediction model, employing a multi-task joint learning strategy, optimizing the parameters of the initial performance prediction model through a gradient descent algorithm to minimize the overall loss function, and determining the performance prediction model based on the optimized model parameters. Specifically, firstly, based on the complete method flow of this invention, multiple known copper ferrite denitration catalyst samples from historical batches are processed, wherein, for each sample, based on its microscopic image set before microwave processing and the set microwave field parameters, the corresponding process structure feature tensor is calculated using the method, serving as the input feature of the sample; simultaneously, the true performance of the sample after processing with the corresponding microwave field parameters is measured through standard catalyst performance testing experiments, serving as the true performance label of the sample; a large number of such samples (each sample containing a process structure feature tensor and a set of corresponding true performance labels) together constitute the training set for model training. Subsequently, the training set is input into the defined initial performance prediction model (whose structure includes a spatiotemporal feature encoder, a multi-head attention fusion layer, and a regression prediction layer). A multi-task joint learning strategy is adopted, whereby the model's regression prediction layer simultaneously outputs two predicted values: denitrification efficiency and stability index. An overall loss function is constructed for this purpose, which is a weighted sum of individual task losses, such as the mean squared error loss (used for regression prediction). Gradient descent algorithms (such as the Adam optimizer) and their variants are used to iterate multiple times on the training set. The gradient of the loss function with respect to all trainable parameters (such as weights and biases) is calculated through backpropagation, and the parameters are updated in the opposite direction of the gradient to minimize the overall loss function. Early stopping on a validation set can be used during training to prevent overfitting. Finally, when the model's performance on the validation set stabilizes and reaches its optimum, training is stopped, and the optimized model parameters are fixed. The resulting model with stable predictive capabilities is the completed performance prediction model.

[0065] It should be noted that the overall loss function It consists of the weighted sum of the losses from each task: ; In the formula, and α and β represent the losses from predicting denitrification efficiency and stability index, respectively (usually expressed as mean squared error, MSE), while α and β are hyperparameters balancing the importance of the two tasks. The optimizer (e.g., Adam) minimizes these losses using gradient descent. To update all parameters of the model.

[0066] It should be noted that the catalyst performance test experiment is to test its nitrogen oxide conversion rate in a denitrification reactor under fixed conditions to obtain the denitrification efficiency, and to evaluate its activity decay degree through long-term operation or cycle test to obtain the stability index.

[0067] Please refer to Figure 4 In some embodiments, the performance prediction model includes a spatiotemporal feature encoder, a multi-head attention fusion layer, and a regression prediction layer connected in sequence. The step of inputting the process structure feature tensor into the pre-trained performance prediction model to obtain the predicted performance results of the copper ferrite denitration catalyst after treatment under the microwave field parameters includes steps S401 to S403: Step S401: Input the process structure feature tensor into the pre-trained performance prediction model, and extract the local structural features and spatiotemporal dependencies in the process structure feature tensor through the spatiotemporal feature encoder to construct the encoding result; In some embodiments, the process structure feature tensor (assuming its dimensions are [Batch, Channels, Features], where Features integrate spatial and trend dimensions) is input into a spatiotemporal feature encoder. This encoder first extracts structural features within the local neighborhood through a one-dimensional convolutional layer (CNN). Subsequently, the feature sequence output by the convolutional layer is reshaped along the feature dimensions and input into a Long Short-Term Memory (LSTM) network layer or a self-attention module. Taking self-attention as an example, its core computation is as follows: In this context, the query (Q), key (K), and value (V) are all obtained by linear projection of the input sequence. is the dimension of the key vector, where this process models long-range dependencies between features and outputs an encoded result that integrates local and global contextual information.

[0068] Step S402: The encoding result is processed by the multi-head attention fusion layer to obtain a dynamic evolution trend vector; In some embodiments, the encoded result is input to a multi-head attention fusion layer, which linearly projects the encoded result into multiple independent query, key, and value vectors. By computing multiple attention heads in parallel, the model can focus on different parts of the encoded result from different representation subspaces; for example, one attention head might focus on the evolutionary correlation of pore structure, while another focuses on the variation pattern of particle size. The outputs of these heads are then concatenated and projected again to form a unified, deeply fused feature vector.

[0069] It should be noted that the calculation process of the multi-head attention fusion layer is as follows: Bullish Attention: ; Each attention head: ; In the formula, , , and is a learnable projection weight matrix, and h is the number of attention heads. This layer performs parallel multi-view information focusing and fusion on the encoded results, outputting a deeply abstract dynamic evolution trend vector.

[0070] It should be noted that the feature dimension of the input encoding result is . Then the key and value dimensions of each attention head are usually set to This ensures that the total computational cost of the multi-head attention mechanism is moderate and that it can effectively capture the representational information of different subspaces.

[0071] It should be noted that the dynamic evolution trend vector integrates all the key information and their complex interactions in the encoding result, and is called the dynamic evolution trend vector. It is a deep abstract representation of the final evolution state of the process structure under microwave action.

[0072] Step S403: The dynamic evolution trend vector is nonlinearly mapped through the regression prediction layer to output the predicted performance results; wherein, the predicted performance results include denitrification efficiency and catalyst stability index.

[0073] In some embodiments, the dynamic evolution trend vector is input to the final layer of the performance prediction model—the regression prediction layer. This layer typically consists of one or more fully connected layers and uses a nonlinear activation function such as ReLU. It maps the high-dimensional dynamic evolution trend vector to specific performance scalar values ​​through a series of nonlinear transformations. This layer ultimately outputs two predicted values: one is the denitrification efficiency (usually expressed as a percentage, such as nitrogen oxide conversion rate), and the other is a catalyst stability index (e.g., expressed as activity retention rate or decay coefficient). These two output values ​​together constitute the predicted performance result of the copper ferrite denitrification catalyst after treatment under specified microwave field parameters.

[0074] By inputting the process structure feature tensor into the trained performance prediction model, the model can make inferences based on the deeply integrated features that reflect the entire process mechanism, thereby achieving a more accurate and reliable prediction of the final performance.

[0075] This invention establishes an accurate predicted material starting point by acquiring a set of microscopic images before microwave processing and extracting the initial structural feature vector, avoiding prediction bias caused by ignoring differences in the initial structure. Based on microwave field parameters, a local field strength distribution feature vector corresponding to the microscopic structure space is determined. This method accurately characterizes the non-uniform physical nature of the interaction between microwaves and catalysts, transforming macroscopic microwave parameters into real local driving conditions affecting structural evolution, fundamentally overcoming model errors caused by treating the microwave field as a uniform field. The initial structural evolution trend vector is obtained by matching the microwave field parameters from a pre-constructed microwave structural evolution mapping library. This method introduces universal dynamic laws of structural evolution under specific microwave parameters, revealed by historical experimental data, making prediction no longer limited to a single final static structure. Using the local field strength distribution feature vector reflecting spatial differences as conditional weights, the initial structural evolution trend vector is adjusted to obtain the target structural evolution trend vector. This innovative step achieves intelligent fusion of "universal dynamic laws" and "current sample-specific local conditions," enabling the prediction model to dynamically adapt to actual evolution paths under different microscopic environments. By fusing the target structure evolution trend vector with the initial structure feature vector, a process structure feature tensor is obtained. This tensor integrates the initial state, dynamic evolution law, and local field coupling effect, forming a high-dimensional information carrier that comprehensively represents the "microwave process-structure dynamic response." This process structure feature tensor is then input into a trained performance prediction model, enabling the model to make inferences based on the deeply fused features that reflect the entire process mechanism, thereby achieving more accurate and reliable predictions of the final performance.

[0076] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0077] Based on the above embodiments of the data-driven microwave-assisted copper ferrite denitration catalyst performance prediction method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the data-driven microwave-assisted copper ferrite denitration catalyst performance prediction method of any embodiment of the present invention.

[0078] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0079] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0080] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0081] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the data-driven microwave-assisted copper ferrite denitration catalyst performance prediction method described in any of the above-described method embodiments of the present invention.

[0082] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0083] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalysts, characterized in that, include: Obtain a set of microscopic images of the copper ferrite denitration catalyst before microwave treatment; Based on preset microwave field parameters, a local field strength distribution feature vector corresponding to the microstructure space of the copper ferrite denitration catalyst is determined. An initial structural feature vector of the copper ferrite denitration catalyst is extracted from the microscopic image set and matched with a pre-constructed microwave structure evolution mapping library according to the microwave field parameters to obtain a corresponding initial structural evolution trend vector. The initial structural evolution trend vector is adjusted with the local field strength distribution feature vector as a conditional weight to obtain a target structural evolution trend vector. The target structural evolution trend vector is then fused with the initial structural feature vector to obtain a process structural feature tensor reflecting the interaction between the microwave field and the catalyst structure. The microwave structure evolution mapping library is obtained by processing time-series structural image sequences under different microwave field parameters. The process structure feature tensor is input into a pre-trained performance prediction model to obtain the predicted performance results of the copper ferrite denitration catalyst after treatment under the microwave field parameters.

2. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 1, characterized in that, The performance prediction model includes a spatiotemporal feature encoder, a multi-head attention fusion layer, and a regression prediction layer connected in sequence. The process structure feature tensor is input into the pre-trained performance prediction model to obtain the predicted performance results of the copper ferrite denitration catalyst after treatment under the microwave field parameters, including: The process structure feature tensor is input into a pre-trained performance prediction model, and the local structural features and spatiotemporal dependencies in the process structure feature tensor are extracted by the spatiotemporal feature encoder to construct the encoding result. The encoding result is processed by the multi-head attention fusion layer to obtain a dynamic evolution trend vector; The regression prediction layer performs a nonlinear mapping on the dynamic evolution trend vector to output the predicted performance results; wherein, the predicted performance results include denitrification efficiency and catalyst stability index.

3. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 1, characterized in that, The determination of the local field strength distribution characteristic vector corresponding to the microstructure space of the copper ferrite denitration catalyst based on preset microwave field parameters includes: The microwave field parameters are normalized to obtain normalized microwave parameters, wherein the microwave field parameters include microwave frequency, microwave power and microwave irradiation time. Based on the normalized microwave parameters, a reference field strength vector reflecting the average field strength level is calculated. Based on the porosity and composition distribution information reflected in the microscopic image set, the dielectric constant distribution feature vector at each location in the microstructure space is estimated through finite element electromagnetic simulation. The local field strength distribution feature vector is generated by performing element-wise multiplication between the reference field strength vector and the dielectric constant distribution feature vector.

4. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 3, characterized in that, The step of performing element-wise multiplication of the reference field strength vector and the dielectric constant distribution feature vector to generate the local field strength distribution feature vector includes: Calculate the spatial autocorrelation coefficient of the dielectric constant distribution eigenvector; The dielectric constant distribution feature vector is weighted and smoothed based on the spatial autocorrelation coefficient to generate a correlation feature vector that reflects spatial dependence. The reference field strength vector and the associated feature vector are multiplied element-wise to generate the local field strength distribution feature vector.

5. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 1, characterized in that, The extraction of the initial structural feature vector of the copper ferrite denitration catalyst from the microscopic image set includes: The microscopic image set is preprocessed to obtain a standardized microscopic image set, wherein the microscopic image set includes scanning electron microscope images and transmission electron microscope images; Based on image processing algorithms, multidimensional morphological features are extracted from the standardized microscopic image set to form the initial structural feature vector of the copper ferrite denitration catalyst; wherein, the morphological features include average particle size, size distribution variance, porosity and shape factor.

6. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 1, characterized in that, The step of matching the microwave field parameters with a pre-built microwave structure evolution mapping library to obtain the corresponding initial structure evolution trend vector includes: The microwave field parameters are matched with the index keys pre-stored in the microwave structure evolution mapping library to obtain the matching results; Based on the matching results, the associated temporal structure image sequence is determined; Calculate the difference vector sequence of structural features between adjacent time frames from the temporal structure image sequence; The difference vector sequence is fitted with a time series, and the feature parameters corresponding to the changing trend are extracted to form the initial structural evolution trend vector.

7. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 6, characterized in that, The microwave structure evolution mapping library is obtained by training time-series structure image sequences under different microwave field parameters, including: Using several combinations of initial microwave field parameters, the initial copper ferrite denitration catalyst samples were microwave-treated, and the initial time-series structure image sequence changing over time during the treatment process was acquired. The initial temporal structure image sequence is subjected to image denoising and feature alignment to obtain the processing result; Each of the initial microwave field parameters is combined as an index key, and the corresponding processing result is used as a data value to establish a one-to-one mapping relationship. Based on all the established mapping relationships, a microwave structure evolution mapping library is generated using the index key as the retrieval basis.

8. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 1, characterized in that, The step of adjusting the initial structure evolution trend vector using the local field strength distribution feature vector as conditional weights to obtain the target structure evolution trend vector includes: Based on the local field strength distribution feature vector, calculate the weight distribution of the initial structure evolution trend vector; The initial structural evolution trend vector is weighted and fused according to the weight distribution to obtain the target structural evolution trend vector.

9. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 1, characterized in that, The step of fusing the target structure evolution trend vector with the initial structure feature vector to obtain the process structure feature tensor includes: Perform a tensor concatenation operation on the target structure evolution trend vector and the initial structure feature vector to obtain a concatenated feature tensor; The spliced ​​feature tensor is fused to output the process structure feature tensor.

10. The data-driven method for predicting the performance of microwave-assisted copper ferrite denitration catalyst according to claim 1, characterized in that, The training process of the performance prediction model includes: Obtain a training set, wherein the training set includes several samples, each sample including an initial process structure feature tensor and a corresponding true performance label; The training set is input into the initial performance prediction model. A multi-task joint learning strategy is adopted, and the parameters of the initial performance prediction model are optimized by the gradient descent algorithm to minimize the overall loss function. The performance prediction model is determined based on the optimized model parameters.

Images