Catalyst Activity Analysis Method and System Based on Material Surface Physicochemical Indicators

Abstract

This application provides a method and system for catalyst activity analysis based on material surface physicochemical indicators, relating to the field of water treatment technology. The method first acquires a sample dataset, divides it into a training set and a test set, trains multiple candidate models using the training set, and then selects a prediction model from the candidate models. The catalyst data to be tested is then input into the prediction model to obtain the predicted degradation rate output by the model. The sample dataset includes surface physical property parameters such as pore depth and grayscale difference of the catalyst, and surface chemical property parameters such as the peak height ratio of functional groups. The method can train candidate models using Bayesian optimization and select a prediction model by combining evaluation indicators calculated from the test set. This allows for the use of multi-model fusion data to enhance the machine learning process in environments with small sample machine learning data, thereby improving the prediction accuracy and generalization ability of catalyst activity analysis.

Description

Technical Field

[0001] This application relates to the field of water treatment technology, and in particular to a method and system for analyzing catalyst activity based on the physicochemical indicators of material surfaces. Background Technology

[0002] Environmental catalytic oxidation technology is a technique that uses catalysts to promote oxidation reactions and efficiently degrade organic pollutants in the environment. Environmental catalytic oxidation technology can be applied in the field of water treatment; for example, it can be used in advanced oxidation processes (AOPs) based on peroxymonosulfate (PMS) to degrade recalcitrant organic pollutants in water bodies.

[0003] Since the core of environmental catalytic oxidation technology lies in the catalyst, and the catalyst's activity and stability directly determine the treatment effect and cost, the implementation of environmental catalytic oxidation technology requires analysis of the catalyst's activity, stability, and other properties. However, catalyst performance is influenced by a complex array of factors. Therefore, analyzing catalyst performance necessitates conducting batch-based repeated experiments based on different factor indicators, testing the degradation efficiency under experimental conditions, and then combining this with data analysis methods such as regression analysis to predict the optimal factor indicators. Only after successful experimental verification can the catalyst be deployed in practical applications.

[0004] However, since the performance analysis of catalysts relies heavily on repeated and tedious batch experiments to test degradation efficiency, the catalyst activity analysis method based on test experiments is time-consuming, laborious and costly. Therefore, it is difficult to establish accurate quantitative structure-activity relationship (QSAR), resulting in insufficient prediction accuracy and generalization ability of the catalyst activity analysis process. Summary of the Invention

[0005] In view of this, embodiments of this application provide a method and system for catalyst activity analysis based on material surface physicochemical indicators, in order to solve the problems of insufficient prediction accuracy and generalization ability in the catalyst activity analysis process.

[0006] According to a first aspect of this application, a method for analyzing catalyst activity based on physicochemical indicators of material surfaces is provided, the method comprising:

[0007] A sample dataset is obtained, which consists of historical experimental data tagged with pollutant degradation rates. The sample dataset includes surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters of the catalyst. The surface physical property parameters include pore depth grayscale difference extracted from scanning electron microscope images. The surface chemical property parameters include functional group peak height ratio extracted from infrared spectra.

[0008] The sample dataset is divided into a training set and a test set;

[0009] Multiple candidate models are trained using the training set. The candidate models are machine learning models obtained by searching for the optimal hyperparameter combination using a Bayesian optimization algorithm with the pollutant degradation rate as the prediction target. The multiple candidate models include at least a random forest model, an extreme gradient boosting model, and a stacked ensemble model.

[0010] A prediction model is selected from the candidate models. The prediction model is the optimal candidate model determined based on the evaluation metrics determined by evaluating the performance of multiple trained candidate models using the test set. The evaluation metrics include at least one of the coefficient of determination, mean absolute error, and root mean square error.

[0011] Acquire the catalyst data to be tested, and input the catalyst data to be tested into the prediction model to obtain the predicted degradation rate output by the prediction model.

[0012] In some embodiments, obtaining a sample dataset includes:

[0013] Obtain application information of the catalyst to be tested;

[0014] Based on the application information, a keyword combination is constructed, which includes catalyst keywords, pollutant keywords, and type keywords;

[0015] Based on the keyword combination, retrieve real experimental data within a preset analysis period from the database;

[0016] The surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters are identified from the real experimental data to obtain the historical experimental data.

[0017] The historical experimental data are labeled with the pollutant degradation rate in the real experimental data to generate the sample dataset.

[0018] In some embodiments, the surface physical property parameters, the surface chemical property parameters, and the catalytic reaction process condition parameters are identified from the real experimental data to obtain the historical experimental data, including:

[0019] Scanning electron microscope images of the catalyst were extracted from the actual experimental data.

[0020] The scanning electron microscope image is converted into a grayscale image, and the hole region is identified in the grayscale image;

[0021] The opening region and the bottom region are determined along the central axis of the hole region. The opening region is an annular region extending inward from the edge of the hole region by a predetermined number of pixels. The bottom region is a square region containing the deepest part of the hole region.

[0022] The grayscale difference between the orifice region and the orifice bottom region is calculated using the image histogram method to obtain the orifice depth grayscale difference.

[0023] In some embodiments, the surface physical property parameters, the surface chemical property parameters, and the catalytic reaction process condition parameters are identified from the real experimental data to obtain the historical experimental data, including:

[0024] The infrared spectrum of the catalyst was extracted from the actual experimental data;

[0025] Read the stretching vibration peak heights from the infrared spectrum. The stretching vibration peak heights include the carbonyl stretching vibration peak height, the single carbon-oxygen stretching vibration peak height, and the ferro-oxygen stretching vibration peak height.

[0026] The functional group peak height ratio is calculated based on the stretching vibration peak height, and the functional group peak height ratio includes a first peak height ratio and a second peak height ratio; the first peak height ratio is the ratio of the carbonyl stretching vibration peak height to the single carbon-oxygen bond stretching vibration peak height; the second peak height ratio is the ratio of the ferrite bond stretching vibration peak height to the single carbon-oxygen bond stretching vibration peak height.

[0027] In some embodiments, the sample dataset is divided into a training set and a test set, including:

[0028] A high-efficiency machine learning regression algorithm based on the gradient boosting framework, combined with the mean method, is used to fill in the missing values ​​in the sample dataset;

[0029] The sample dataset was transformed using the Yeo-Johnson transform to generate a normally distributed dataset;

[0030] The normally distributed dataset is standardized using a normalization scaler to obtain a standardized dataset.

[0031] The standardized dataset is divided into a training set and a test set according to the five-fold cross-validation mechanism.

[0032] In some embodiments, training multiple candidate models using the training set includes:

[0033] The training set is input into multiple candidate models, and the model architectures used by the multiple candidate models are all or partly different.

[0034] Obtain the training prediction results output by the candidate model;

[0035] The training loss is calculated based on the training prediction results and the pollutant degradation rate labels in the training set;

[0036] Iterative training is performed based on the training loss to adjust the model parameters of the candidate model.

[0037] In some embodiments, iterative training is performed based on the training loss to adjust the model parameters of the candidate models, including:

[0038] Initialize the iteration parameters, which include the number of exploration iterations and the number of optimization iterations;

[0039] Iterative training is performed on the candidate model based on the iteration parameters and the training loss; wherein the candidate model is configured to use a Bayesian optimization algorithm to search the hyperparameter space during the iterative training process to determine the optimal combination of hyperparameters of the candidate model.

[0040] The model parameters of the candidate model are output based on the optimal hyperparameter combination.

[0041] In some embodiments, iterative training of the candidate model is performed based on the iteration parameters and the training loss, including:

[0042] A Bayesian optimization algorithm is used, with the average performance under the five-fold cross-validation strategy as the evaluation criterion, to search the hyperparameter space of the random forest model and the extreme gradient boosting model respectively, in order to obtain the optimal hyperparameter combination.

[0043] Based on the optimal hyperparameter combination, the meta-feature vectors of the random forest model and the extreme gradient boosting model are generated;

[0044] Using the meta-feature vector as the new input feature and the pollutant degradation rate in the sample dataset as the output label, a training dataset is constructed.

[0045] A meta-learner is constructed using the minimum absolute shrinkage and selection operator cross-validation algorithm, and the model is trained on the training dataset using the meta-learner to obtain the stacked ensemble model.

[0046] In some embodiments, selecting a prediction model from the candidate models includes:

[0047] The test set is input into the trained candidate models;

[0048] Obtain the test prediction results output by the candidate models;

[0049] Evaluation indicators are calculated based on the test prediction results and the pollutant degradation rate labels in the test set;

[0050] Based on the determination coefficient, the prediction accuracy of the candidate models is evaluated to determine the high-accuracy models whose determination coefficient is higher than a preset coefficient threshold.

[0051] The high-precision model is validated using the mean absolute error and the root mean square error to determine the optimal alternative model.

[0052] According to a second aspect of this application, a catalyst activity analysis system based on material surface physicochemical indicators is provided, the system comprising:

[0053] The sample data acquisition module is used to acquire a sample dataset, which is historical experimental data tagged with pollutant degradation rate. The sample dataset includes surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters of the catalyst. The surface physical property parameters include pore depth grayscale difference extracted from scanning electron microscope images. The surface chemical property parameters include functional group peak height ratio extracted from infrared spectra.

[0054] The dataset partitioning module is used to divide the sample dataset into a training set and a test set;

[0055] The model training module is used to train multiple candidate models using the training set. The candidate models are machine learning models obtained by searching for the optimal hyperparameter combination using a Bayesian optimization algorithm with the pollutant degradation rate as the prediction target. The multiple candidate models include at least a random forest model, an extreme gradient boosting model, and a stacked ensemble model.

[0056] A model selection module is used to select a prediction model from the candidate models. The prediction model is the optimal candidate model determined based on the performance evaluation of multiple trained candidate models using the test set and the evaluation metrics. The evaluation metrics include at least one of the coefficient of determination, mean absolute error, and root mean square error.

[0057] The prediction module is used to acquire data of the catalyst to be tested and input the data of the catalyst to be tested into the prediction model to obtain the predicted degradation rate output by the prediction model.

[0058] According to a third aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described method for analyzing catalyst activity based on physicochemical indicators of material surfaces.

[0059] According to a fourth aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described method for analyzing catalyst activity based on physicochemical indicators of material surfaces.

[0060] By employing the above technical solution, embodiments of this application provide a method and system for catalyst activity analysis based on material surface physicochemical indicators. The method first acquires a sample dataset, divides it into a training set and a test set, trains multiple candidate models using the training set, and then selects a prediction model from the candidate models. The catalyst data to be tested is then input into the prediction model to obtain the predicted degradation rate output by the prediction model. The sample dataset includes surface physical property parameters such as pore depth and grayscale difference of the catalyst, as well as surface chemical property parameters such as the peak height ratio of functional groups. This method can train candidate models using Bayesian optimization and select a prediction model by combining evaluation indicators calculated from the test set. It can enhance the machine learning process using multi-model fusion data in environments with small sample machine learning data volumes, thereby improving the prediction accuracy and generalization ability of catalyst activity analysis.

[0061] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0062] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0063] Figure 1 A schematic diagram of the catalyst activity analysis method based on material surface physicochemical indicators provided in this application embodiment;

[0064] Figure 2 This is a schematic diagram of the catalyst product testing and analysis workflow provided in the embodiments of this application;

[0065] Figure 3 A schematic diagram of the Pearson correlation heatmap for the features provided in the embodiments of this application;

[0066] Figure 4This is a schematic diagram illustrating the performance evaluation results of different models provided in the embodiments of this application;

[0067] Figure 5 This is a schematic diagram illustrating the effect of the determination coefficient on the catalytic material surface property dataset provided in the embodiments of this application;

[0068] Figure 6 This is a schematic diagram illustrating the model accuracy of the RF model with different dataset partitioning methods provided in the embodiments of this application;

[0069] Figure 7 A schematic diagram of the structure of a catalyst activity analysis system based on material surface physicochemical indicators provided in this application embodiment. Detailed Implementation

[0070] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0071] In this application, a catalyst refers to a material used in environmental catalytic oxidation technology, which is a reactant with specific surface physical and chemical properties. In some embodiments, the catalyst may include metal catalysts, metal oxide catalysts, metal complex catalysts, heteropolyacid acid salt catalysts, biocatalysts, nitrogen oxide catalysts, and photocatalysts, etc.

[0072] Catalysts can promote oxidation reactions in the environment, thereby efficiently degrading organic pollutants. In some embodiments, catalysts can be applied in conjunction with environmental catalytic oxidation technologies in the field of water treatment, for example, in advanced oxidation processes (AOPs) based on peroxymonosulfate (PMS) to degrade recalcitrant organic pollutants in water.

[0073] Since the core of environmental catalytic oxidation technology lies in the catalyst, and the catalyst's activity and stability directly determine the treatment effect and cost, it is necessary to analyze the catalyst's activity, stability, and other properties when implementing environmental catalytic oxidation technology.

[0074] Catalyst performance is influenced by a complex array of factors. In some embodiments, analyzing catalyst performance requires conducting batch-repeated experiments based on different factor indices, testing degradation efficiency under experimental conditions, and then combining data analysis methods such as regression analysis to predict optimal factor indices. After experimental verification, the catalyst is then deployed in practical applications.

[0075] However, since catalyst performance analysis heavily relies on repeated and tedious batch experiments to test degradation efficiency, catalyst activity analysis based on test experiments is time-consuming, labor-intensive, and costly, making it difficult to establish accurate quantitative structure-activity relationship (QSAR). Therefore, in some embodiments, rapid physicochemical index catalyst activity analysis can also be performed based on machine learning (ML) technology.

[0076] In the process of catalyst activity analysis based on machine learning technology, experimental data can be collected first, and then the data can be cleaned, preprocessed, and labeled to obtain training sample data. The training sample data is then used to train a machine learning model, which is then used to predict catalyst activity.

[0077] However, in the field of catalysts, the amount of experimental data that can meet the needs of machine learning is scarce, especially for environmental catalytic oxidation processes of carbon-based and persulfate (PMS) systems. This creates a small sample environment with limited available experimental data, many feature dimensions, and complex correlations between indicators, resulting in insufficient model prediction accuracy and generalization ability.

[0078] To address the issues of insufficient prediction accuracy and generalization ability in catalyst activity analysis, this application provides a catalyst activity analysis method based on material surface physicochemical indicators in some embodiments. This method integrates scanning electron microscopy (SEM) image indicators of catalytic materials with infrared (IR) spectral data, and correlates them with the prediction of pollutant degradation activity of environmental catalysts. By training a machine learning model on small sample data, the physical and chemical parameters of the material are fused, forming a rapid physicochemical indicator-based catalyst activity analysis method based on machine learning. This addresses the problems of insufficient model prediction accuracy and generalization ability, and alleviates issues such as numerous steps, long processing times, disconnect between preparation and testing processes, and heavy reliance on experience in catalyst activity analysis.

[0079] The method can be applied to electronic devices with data processing capabilities. These electronic devices include, but are not limited to, computers, servers, mobile terminals, smart wearable devices, and industrial control machines. For ease of description, this application embodiment uses an electronic device as the execution subject of the method. It should be understood that the method can also be applied to other types of execution subjects, which are not illustrated in this application embodiment. Figure 1 As shown, the method includes:

[0080] S101. Obtain the sample dataset.

[0081] When performing catalyst activity analysis, a sample dataset needs to be acquired first for model training. This sample dataset consists of historical experimental data tagged with pollutant degradation rates. Depending on the needs of the catalyst activity analysis process, the sample dataset may include surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters of the catalyst.

[0082] For example, the surface physical properties of a catalyst include particle diameter, pore size, pore depth, and porosity. Surface chemical properties may include the concentration of carbonyl (C=O) functional groups, the concentration of single-bonded carbon-oxygen (CO) functional groups, and the height of ferro-oxygen bonds (Fe-O). Catalytic reaction process conditions may include catalyst dosage, initial contaminant concentration, persulfate dosage, and pH.

[0083] In some embodiments, the sample dataset can be generated from real experimental data. Specifically, when acquiring the sample dataset, application information of the catalyst to be tested can be obtained, and keyword combinations can be constructed based on this information. Then, real experimental data within a preset analysis period can be retrieved from the database based on these keyword combinations. The keyword combinations include catalyst keywords, pollutant keywords, and type keywords.

[0084] For example, focusing on the activation of persulfate degradation of pollutants in water by carbon-based catalysts, the keyword combination was set as carbon catalysis, pollutant keywords as BP or BPA, and type keyword as PMS. Peer-reviewed research data from the past 8 years were retrieved from databases such as Science Direct and Google Scholar, and 112 real experimental data were collected.

[0085] Next, surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters are identified from real experimental data to obtain historical experimental data. Then, the historical experimental data are labeled with the pollutant degradation rate in the real experimental data to generate a sample dataset.

[0086] For example, considering the diversity of data sources, the surface physical properties, surface chemical parameters, and catalytic process conditions of the catalyst in the catalytic reaction system can be identified as influencing variables, and the catalytic activity can be labeled with the degradation rate (Dr), thus constructing a sample dataset containing 639 samples.

[0087] In some embodiments, surface physical property parameters can be obtained through image recognition of SEM images of the catalyst. For example, surface particle diameter, pore size, pore depth, porosity, and other surface physical property parameters can be obtained by processing the SEM image into a grayscale image using ImageJ software, followed by threshold segmentation and grayscale histogram analysis.

[0088] In the preparation and application of environmental catalysts, the depth of pores on the catalyst surface affects the diffusion efficiency of reactants and products within the catalyst channels. Deeper pores imply longer mass transfer paths, leading to decreased utilization of internal active sites. Furthermore, pore depth is related to the three-dimensional structure of the catalyst surface; it can indirectly reflect the accessibility of active sites and also influence mass transfer within the catalyst.

[0089] Therefore, the physical surface characteristics of environmental catalysts can include not only two-dimensional planar information such as pore size distribution and overall porosity, but also three-dimensional structural features of the pores, such as structural features along the pore depth. Pore depth directly affects the mass transfer efficiency of reactants and the availability of internal active sites. By defining a quantitative index of pore depth grayscale, the three-dimensional structural features of the pores are extracted from SEM images, serving as a key three-dimensional supplementary information for evaluating the catalyst's microstructure. That is, the surface physical property parameters can also include the pore depth grayscale difference extracted from scanning electron microscope images.

[0090] In some embodiments, when performing the process of identifying surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters from real experimental data to obtain historical experimental data, scanning electron microscope images of the catalyst can be extracted from the real experimental data, the scanning electron microscope images can be converted into grayscale images, and pore regions can be identified in the grayscale images.

[0091] Next, the opening region and the bottom region of the hole are determined along the central axis of the hole region. The opening region is a ring-shaped area extending inwards from the edge of the hole region by a predetermined number of pixels; the bottom region is a square area containing the deepest part of the hole region. Then, the grayscale difference between the opening region and the bottom region is calculated using the image histogram method to obtain the hole depth grayscale difference.

[0092] For example, such as Figure 2As shown, secondary electron images of the catalyst sample can be obtained through testing and processed into 8-bit grayscale images using scanning electron microscope (SEM) software. The pore opening and pore bottom regions are defined along the central axis of the pore area. The pore opening region is defined as a ring-shaped area extending inward from the pore edge for 3-5 pixels. The pore bottom region is defined as the square region at the deepest point of the pore. The grayscale difference between the pore opening and pore bottom regions is calculated using the image histogram method; the grayscale difference ranges from 2 to 7 for different catalysts. To obtain more accurate surface physical property parameters, multiple multi-point measurements can be performed, and the average value of the measurement results is used as the output.

[0093] In some embodiments, surface chemical property parameters can be obtained by spectroscopic analysis of the IR spectrum of the catalyst. For example, surface chemical property parameters are expressed as specific peaks in the IR pattern, such as 3420 cm⁻¹. - ¹OH at the location; 1620 cm - C=O at ¹; 1380 cm - CO at ¹; 570 cm - After determining the peak height of Fe-O at position ¹, the peak height ratios are obtained by comparison.

[0094] In carbon-based catalysis, the persulfate-activated catalysis process can form a quinone and hydroquinone cycle. When a C=O group is adjacent to a CO group, they interact through electronic effects on the carbon skeleton. This fine-tuning of the electronic structure directly affects the adsorption strength and activation ability of persulfate (PMS) and other reactant molecules at that site. In this cycle, C=O (quinone group) and C-OH (phenolic hydroxyl group) interconvert, jointly achieving electron transfer, thereby efficiently activating PMS to generate sulfate radicals. Therefore, when optimizing carbon-based catalysts, the absolute concentration of a specific functional group (such as C=O) on the surface can be increased.

[0095] Catalytic activity is not determined independently by a single functional group, but strongly depends on the synergy and balance between different functional groups, such as the synergy between C=O, CO, and Fe-O. To improve catalyst performance stability and preparation reproducibility, two core indicators for evaluating catalyst surface chemical properties can be defined: the functional group peak-height ratio, including a first peak-height ratio and a second peak-height ratio. The first peak-height ratio is the ratio of the carbonyl stretching vibration peak height to the single-bond carbon-oxygen stretching vibration peak height, i.e., the C=O / CO peak-height ratio (C / D); the second peak-height ratio is the ratio of the ferro-oxygen bond stretching vibration peak height to the single-bond carbon-oxygen bond stretching vibration peak height, i.e., the Fe-O / CO peak-height ratio (Fe / C). For example, the C=O / CO peak-height ratio ranges from 17.1 to 104.98; the Fe-O / CO peak-height ratio ranges from 4.19 to 79.33.

[0096] The surface chemical properties parameters include the peak height ratio of functional groups extracted from infrared spectra. The relative concentration ratios of C=O, CO, and Fe-O functional groups are used as key indicators for evaluating and guiding catalyst preparation. This ratio more fundamentally reflects the electronic structure and catalytic potential of the catalyst surface, superior to the absolute concentration of any single functional group.

[0097] In some embodiments, when performing the identification of surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters from real experimental data to obtain historical experimental data, the infrared spectrum of the catalyst can be extracted from the real experimental data first. The stretching vibration peak heights are read from the infrared spectrum, and the functional group peak height ratios are calculated based on these peak heights. These stretching vibration peak heights include the carbonyl stretching vibration peak height, the single-bond carbon-oxygen bond stretching vibration peak height, and the ferro-oxygen bond stretching vibration peak height.

[0098] For example, for the first peak height ratio (C=O / CO peak height ratio), it can be read from the infrared spectrum of the catalyst at 1620 cm⁻¹. - ¹The peak height of the C=O stretching vibration near H_C=O, and 1380 cm⁻¹ - The peak height of the CO stretching vibration near ¹ (HC-O) is HC=O / H_C-O. The first peak height ratio reflects the relative density of carbonyl active sites on the catalyst surface relative to the basic carbon-oxygen skeleton, and is a key parameter for measuring the degree of surface oxidation and catalytic potential.

[0099] For the second peak height ratio (Fe-O / CO peak height ratio), the 570 cm⁻¹ peak in the catalyst's infrared spectrum can be read. -The ratio of the characteristic peak height of Fe-O (HFe-O) near ¹ to the peak height of CO (HC-O) mentioned above is HFe-O / HC-O. This second peak height ratio can reflect the relative intensity between the doped iron species and the carbon support, and can be used to assess the dispersion of the metal sites and the strength of their interaction with the support.

[0100] The catalytic reaction process parameters can be set based on different data requirements, and can affect the reaction judgment criteria and algorithm judgment criteria. For example, a reaction can be performed for 20 minutes with 2 grams of catalyst, 200 mg / L PMS oxidant, and 50 mg / L BPA concentration. The reaction judgment criterion is whether the degradation rate reaches 90%; the algorithm judgment criterion is that the difference between the predicted and measured catalytic activity results is within 10%.

[0101] S102. Divide the sample dataset into a training set and a test set.

[0102] After obtaining the sample dataset, it can be divided into a training set and a test set. The training set is used to train the model, and therefore can constitute the majority of the entire sample dataset. The test set is used to evaluate the performance of the trained model, and thus can constitute a minority of the entire sample dataset. For example, the ratio of training to test set data can be 7:3.

[0103] In some embodiments, when dividing the sample dataset into training and test sets, data preprocessing can be performed. This involves using an efficient machine learning regression algorithm based on a gradient boosting framework combined with the mean method to impute missing values ​​in the sample dataset, and then using the Yeo-Johnson transform to transform the sample dataset to generate a normally distributed dataset. A normalization scaler is then used to standardize the normally distributed dataset to obtain a standardized dataset. Finally, the standardized dataset is divided into training and test sets according to the five-fold cross-validation mechanism.

[0104] For example, during data preprocessing, missing values ​​are imputed using an efficient machine learning regression algorithm based on the gradient boosting framework (LightGBM, LGBM) combined with the mean method. The LGBM regression algorithm can use histograms to bin the data, thereby reducing memory usage and computation time. Furthermore, the LGBM regression algorithm can predict missing values ​​based on feature relationships, while the mean method maintains the statistical properties of the data.

[0105] After imputing missing values ​​in the sample dataset, it can be further processed by applying the Yeo-Johnson transform to ensure a normal distribution, and by standardizing it using the Standard Scaler. The Yeo-Johnson transform is a method to convert data into a distribution closer to normal, transforming skewed data into an approximate normal distribution, which helps improve the performance of certain machine learning models. Standard Scaler standardization is a method to convert data into a distribution with a mean of 0 and a standard deviation of 1. Through the Yeo-Johnson and Standard Scaler transforms, the dimensions of the data can be eliminated, ensuring the sample dataset conforms to a normal distribution, thereby improving data quality and model convergence speed.

[0106] Five-fold cross-validation is a model evaluation method used to assess the performance of machine learning models. It involves dividing the dataset into multiple subsets (e.g., five subsets), using four subsets for training, and the remaining subset for validation. This process is repeated five times, each time selecting a different subset as the validation set. The average performance metric is then used to evaluate the model. Therefore, using five-fold cross-validation to divide the sample dataset into training and test sets can reduce evaluation bias caused by the randomness of data partitioning. The ratio of training to test sets can be 9:1, 8:2, 3:1, 7:3, etc.

[0107] S103. Train multiple candidate models using the training set.

[0108] After dividing the sample dataset into training and test sets, the training set can be used for model training, i.e., multiple candidate models can be trained using the training set. These candidate models are machine learning models obtained by searching for the optimal hyperparameter combination using a Bayesian optimization algorithm, with the pollutant degradation rate as the prediction objective. Furthermore, the multiple candidate models include at least random forest models, extreme gradient boosting models, and stacked ensemble models, and the model architectures used by the multiple candidate models are all or partially different.

[0109] For example, all the candidate models are built based on the scikit-learn library. The candidate models can be linear regression models, multiple regression models, random forest (RF), extreme gradient boosting (XGBoost) models, stacking (Stack) models, support vector regression (SVR), multilayer perceptron (MLP) models, convolutional neural network (CNN) models, and stacked ensemble models composed of multiple architectures from these models.

[0110] In some embodiments, when training multiple candidate models using a training set, the training set can be first input into each candidate model, and the training prediction results output by the candidate models can be obtained. Then, the training loss is calculated based on the training prediction results and the pollutant degradation rate labels in the training set, and iterative training is performed based on the training loss to adjust the model parameters of the candidate models.

[0111] For example, three different types of models can be selected as candidate models: Support Vector Regression (SVR), Multilayer Perceptron (MLP), and Random Forest (RF). By inputting samples containing the training set into the candidate models, we can obtain the predicted degradation rate output by each model: the predicted degradation rate y_pred_svr for the SVR model, the predicted degradation rate y_pred_mlp for the MLP model, and the predicted degradation rate y_pred_rf for the RF model.

[0112] Next, loss functions such as Mean Squared Error (MSE) are selected to calculate the training loss, i.e., SVR loss loss_svr = MSE(y_train, y_pred_svr); MLP loss loss_mlp = MSE(y_train, y_pred_mlp); RF loss loss_rf = MSE(y_train, y_pred_rf). Iterative training is then performed based on the training loss to adjust the model parameters.

[0113] The parameter tuning mechanisms differ across models. For example, when tuning the neural network parameters of an MLP (Multi-Level Processing), gradient descent and backpropagation can be used. The chain rule is used to calculate the gradient of the loss function `loss_mlp` relative to each weight `W` and bias `b` of the MLP model, and then optimizers such as SGD and Adam are used to update the parameters based on the calculated gradient and learning rate. When tuning the parameters of an SVR (Support Vector Redirecting), the SMO (Support Motion Model) algorithm can be used to directly find the optimal model parameters, i.e., the support vectors and their coefficients, by solving a convex optimization problem. When tuning the parameters of a RF (Redirect Resource Framework), multiple decision trees can be constructed for learning. The RF parameters can be adjusted by modifying hyperparameters such as the number of decision trees, maximum depth, and minimum number of samples required for splitting.

[0114] During iterative training, Bayesian optimization algorithms can be used to search the hyperparameter space, determine the optimal hyperparameter combination for each model, and unify the initial number of exploration iterations and the number of optimization iterations to ensure experimental consistency. Specifically, in some embodiments, when performing iterative training based on training loss to adjust the model parameters of candidate models, the iteration parameters can be initialized first, including the number of exploration iterations and the number of optimization iterations.

[0115] Then, based on the iteration parameters and training loss, iterative training is performed on the candidate model to output the model parameters of the candidate model according to the optimal hyperparameter combination. The candidate model is configured to use a Bayesian optimization algorithm to search the hyperparameter space during iterative training to determine the optimal hyperparameter combination of the candidate model.

[0116] Therefore, in some embodiments, when performing iterative training on the candidate model based on the iteration parameters and training loss, a Bayesian optimization algorithm can be used to search the hyperparameter space of the random forest model and the extreme gradient boosting model respectively, using the average performance under the five-fold cross-validation strategy as the evaluation standard, in order to obtain the optimal hyperparameter combination.

[0117] For example, to explore the impact mechanism of data volume on model performance, this study selected three ensemble models—XGBoost, RF, and Stack—to train models using degradation efficiency prediction as the research object. Specifically, a sample dataset (D639) containing 639 data points was input into the three ensemble models for training and prediction, and the changes in model prediction accuracy under the Simpson index were analyzed. The model's generalization performance was evaluated through various dataset partitioning strategies: the training and test sets were partitioned at fixed ratios of 90%:10%, 80%:20%, 70%:30%, and 75%:25%, and the dataset was processed using a five-fold cross-validation method.

[0118] For Stacking ensemble models, the LassoCV algorithm can be used as the base learner to stack and integrate RF and XGBoost models. Its hyperparameters are obtained by combining the hyperparameters of the RF and XGBoost models and then optimizing them using Bayesian methods. The LassoCV algorithm uses default parameter combinations and leverages its feature selection and regularization properties to provide a stable foundation for learning the Stack model.

[0119] When performing Bayesian optimization, the base learner hyperparameters can be optimized and meta-features generated first. That is, the Bayesian optimization algorithm is used, with the average performance under 5-fold cross-validation as the evaluation criterion, to search the hyperparameter space of the random forest model and the extreme gradient boosting model respectively, to obtain their respective optimal hyperparameter combinations.

[0120] Then, based on the optimal hyperparameter combination, meta-feature vectors for the random forest model and the extreme gradient boosting model are generated. For example, using the above optimal hyperparameters and a 5-fold cross-validation strategy, out-of-bag predictions for the random forest model and the extreme gradient boosting model can be generated. Specifically, in each fold, the model is trained using the data from the remaining four folds, and predictions are made for the retained fold; after traversing all folds, the prediction results of each fold are concatenated to form two meta-feature vectors corresponding to the original training set samples, denoted as MRF and MXGB.

[0121] Then, using the meta-feature vector as the new input feature and the pollutant degradation rate in the sample dataset as the output label, a training dataset is constructed. For example, it can be trained and integrated with the model through a meta-learner, using the aforementioned meta-feature vectors MRF and MXGB as new input features and the original degradation rate as the output label to construct a second-layer training dataset.

[0122] The meta-learner is then constructed using minimum absolute shrinkage and selection operator cross-validation algorithms, and the model is trained on the training dataset using the meta-learner to obtain a stacked ensemble model. For example, on the second-layer training dataset, the LassoCV algorithm is used as the meta-learner for training. The LassoCV algorithm uses its default parameter settings, automatically selects the regularization strength using its built-in cross-validation, and utilizes the feature selection properties of L1 regularization to assign optimal weights to the prediction results from the first-layer base learners, thereby forming a robust stacked ensemble model.

[0123] S104. Select a prediction model from the candidate models.

[0124] After training multiple candidate models using the training set, the performance of the trained candidate models can be evaluated using the test set to select a prediction model from the candidate models. When selecting a prediction model from the candidate models, evaluation metrics can be defined when evaluating the performance of the multiple trained candidate models using the test set. These evaluation metrics include the coefficient of determination (R²). 2 The evaluation criteria include at least one of the following: coefficient of determination, mean absolute error (MAE), and root mean square error (RMSE). The optimal candidate model is determined based on these evaluation metrics. Specifically, after the candidate model is trained, the performance of the pollutant degradation efficiency prediction model is evaluated on the test set using the three metrics: coefficient of determination, mean absolute error, and root mean square error, to achieve a comprehensive and objective assessment.

[0125] Therefore, in some embodiments, when selecting a prediction model from candidate models, the test set can be input into multiple trained candidate models first, and then the test prediction results output by the candidate models can be obtained. An evaluation index can then be calculated based on the test prediction results and the pollutant degradation rate labels in the test set. Then, a prediction accuracy evaluation is performed on the candidate models based on the coefficient of determination to determine high-precision models with a coefficient of determination higher than a preset threshold. Furthermore, the high-precision models are validated using mean absolute error and root mean square error to determine the optimal candidate model.

[0126] For example, when selecting a predictive model from candidate models, exploratory data analysis can be performed. This involves using descriptive statistical methods to analyze all features of the sample dataset, understanding the data distribution and value range of each feature. Actual experiments are affected by environmental conditions, resulting in data exhibiting range characteristics; the data distribution and value range are shown in Table 1.

[0127] Table 1. Descriptive statistics of the data;

[0128]

[0129] Wherein, Spd represents the surface particle diameter; a represents the aperture; Hdg represents the pore depth grayscale; P represents the porosity; C / D represents the first peak height ratio (C=O / CO peak-height ratio); Fe / C represents the second peak height ratio (Fe-O / CO peak-height ratio); Cd represents the catalyst dosage; Cc represents the contaminant concentration; Pd represents the persulfate dosage (PMS); pH represents the pH value; Rt represents the reaction time; and Dr represents the degradation rate.

[0130] Pearson correlation heatmaps can be used to quantify and visualize the linear relationships between variables. According to statistical standards, a Pearson correlation coefficient with an absolute value less than 0.3 indicates a weak correlation, while a value greater than 0.7 indicates a high correlation. The results show that the absolute values ​​of the correlation coefficients between all features are less than 0.56, indicating no high linear correlation. Figure 3 As shown. Specific correlation results include: a positive correlation of 0.69 between the C=O / CO peak height ratio and Dr, indicating that higher C=O concentrations result in better treatment effects, as C=O is an active site. A moderate negative correlation of -0.43 between catalyst dosage and pollutant concentration indicates a reactive relationship. A correlation of 0.099 between the Fe-O / CO peak height ratio and Dr suggests that Fe does not play a direct reactive role in the PMS catalytic system, but rather alters the electron distribution of surrounding atoms, which may differ from its role in ozone catalysis. These correlation results are consistent with experimental findings and can provide a basis for analyzing data feature correlations and optimizing machine learning models.

[0131] By comparing the predictive performance of statistical models, machine learning (ML) models, and deep learning models on small-sample multi-modal fusion data, the optimal model can be selected. For example... Figure 4 As shown, the predictive performance of the eight models differs significantly. Among the statistical models, linear regression and multiple regression have the highest average training R-value. 2 The coefficients of determination (R²) for the ML model are 0.236 and 0.375, indicating low prediction accuracy. 2 All of them scored above 0.8, with ensemble models such as Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Stack ensemble scoring above 0.85. These models reduce variance by fusing weak learners and demonstrate the best prediction performance.

[0132] In terms of deep learning models, there is a significant performance difference between Multilayer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs). CNNs struggle to uncover latent features and perform poorly in small-sample scenarios due to insufficient data. MLPs, on the other hand, construct a simple hierarchical architecture through a single feedforward mechanism, reducing model complexity and improving interpretability, and still perform well on small-sample data.

[0133] Therefore, the final test results show that the prediction accuracy of "Extreme Gradient Boosting (XGBoost)" is better than that of Random Forest (RF) and Stacking ensemble models. In the future, the Extreme Gradient Boosting (XGBoost) model can be selected to build test methods.

[0134] In some embodiments, when selecting a prediction model from candidate models, multi-model data fusion can also influence machine learning predictions. For example, the addition of physical and chemical property datasets can improve the prediction accuracy of ML models, with Random Forest (RF) showing the greatest improvement. 2 Rising (0.847-0.906, an increase of 5.7%); the Stack integration model saw a moderate increase (0.837-0.876, an increase of 4.8%), XGboost's R 2 The relatively mild increase (0.831-0.846, an increase of 1.8%) in MAE and RMSE indicators simultaneously confirms this pattern, such as... Figure 5 As shown.

[0135] The results confirm that in small-scale data machine learning, the introduction of physical and chemical property data of catalytic materials plays a role in data stabilization and improvement. By linking the structure and performance of the catalytic material surface, the comprehensive reaction will strengthen the intrinsic correlation of data features, which plays an important role in judging the reaction process of the catalytic system and predicting by machine learning.

[0136] S105. Obtain the data of the catalyst to be tested, and input the data of the catalyst to be tested into the prediction model to obtain the predicted degradation rate output by the prediction model.

[0137] After selecting a prediction model from the candidate models, the selected prediction model can be used to predict the degradation rate, that is, to obtain the data of the catalyst to be tested and input the data of the catalyst to be tested into the prediction model to obtain the predicted degradation rate output by the prediction model.

[0138] For example, during model prediction, for the test samples to be predicted, the hyperparameters of the base learner can be optimized and meta-features generated first. This yields the initial predictions using a random forest model and an extreme gradient boosting model with optimal hyperparameters, resulting in the meta-features MRFtest and MXGBtest for the test set. Subsequently, MRFtest and MXGBtest are input into the trained meta-learner to obtain the final predicted degradation rate.

[0139] like Figure 6 As shown, in the verification of catalyst quality judgment, four random ratio partitions (9:1, 8:2, 3:1, 7:3) and a five-fold cross-validation method can be used based on the training dataset. The XGBoost model's prediction accuracy R under the five partition strategies is shown. 2 All scores exceeded 0.90. This validates the effectiveness of multimodal data in improving model performance. Simultaneously, the trained algorithm is used to establish an activity analysis of catalytic materials, determining the quality of catalyst products under input conditions.

[0140] The data obtained after converting the physical and chemical property test data of the three catalysts into characteristic data and substituting them into the data, and setting the operating conditions as follows: 2 grams of catalyst added, PMS concentration of 200 mg / L, and BPA concentration of 50 mg / L, the predicted and measured catalytic activity results for 20 minutes are shown in Table 2.

[0141] Table 2 Catalytic effect and quality assessment of different catalysts;

[0142]

[0143] As shown in Table 2, the predicted Dr value differs from the actual Dr value by less than 10%, indicating accurate prediction and usability of the model. Using a 90% degradation rate as a benchmark, this model can be used to determine the quality of different production batches of catalyst, thus ensuring stable catalyst production.

[0144] It is evident that judging the quality of environmental catalytic materials can validate the effectiveness of using small-sample machine learning data of the physical and chemical properties of environmental catalysts as multi-mode fusion data to enhance machine learning as the core of analysis. It also examines the correlation between data points during model training. This strategy can improve model accuracy by 5.9%, enhancing prediction accuracy. The methodology established by filling physicochemical property data from SEM and IR tests allows this machine learning model to accurately predict catalyst product quality and provide feedback for catalyst production adjustments. Simultaneously, it provides an efficient data-driven solution and a new operational paradigm for machine learning with small-scale environmental data.

[0145] By applying the technical solutions of the above embodiments, the catalyst activity analysis method based on the physicochemical indicators of material surfaces described in the above embodiments can, in essence, introduce the measured data of the catalyst surface and the test data of surface functional groups into the dataset, establish a machine learning analysis model, strengthen the effective correlation between various dimensions, and improve the model prediction accuracy. A dataset containing 12 key parameters is constructed, and modeling is carried out based on 639 real experimental data. In small-sample ML scenarios, the catalytic activity of the catalyst can be determined through the analysis of the physical and chemical indicators of the catalytic material. Multi-model data fusion can improve the accuracy of prediction. Using this strategy for pre-screening, the amount of data required for optimal prediction accuracy is quantified, and the XGBoost model (R) is adopted. 2 =0.904, MAE=4.583, RMSE=7.395), which can accurately evaluate catalytic performance and improve the efficiency of small-sample ML. Under specified quality control conditions, it can achieve the screening of high-quality catalysts.

[0146] In some embodiments, as a specific implementation of the catalyst activity analysis method based on material surface physicochemical indicators in the above embodiments, some embodiments of this application also provide a catalyst activity analysis system based on material surface physicochemical indicators, such as... Figure 7 As shown, the system includes:

[0147] The sample data acquisition module is used to acquire a sample dataset, which is historical experimental data tagged with pollutant degradation rate. The sample dataset includes surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters of the catalyst. The surface physical property parameters include pore depth grayscale difference extracted from scanning electron microscope images. The surface chemical property parameters include functional group peak height ratio extracted from infrared spectra.

[0148] The dataset partitioning module is used to divide the sample dataset into training and test sets;

[0149] The model training module is used to train multiple candidate models using a training set. The candidate models are machine learning models obtained by searching for the optimal hyperparameter combination using a Bayesian optimization algorithm with the pollutant degradation rate as the prediction target. The multiple candidate models include at least a random forest model, an extreme gradient boosting model, and a stacked ensemble model.

[0150] The model selection module is used to select a prediction model from candidate models. The prediction model is the optimal candidate model determined based on the performance evaluation of multiple trained candidate models using a test set and the evaluation metrics. The evaluation metrics include at least one of the coefficient of determination, mean absolute error, and root mean square error.

[0151] The prediction module is used to acquire data on the catalyst to be tested and to input the data into the prediction model to obtain the predicted degradation rate output by the prediction model.

[0152] By applying the technical solutions of the above embodiments, the catalyst activity analysis system based on material surface physicochemical indicators described in the above embodiments can first acquire a sample dataset through a sample data acquisition module. This sample dataset includes surface physical property parameters such as pore depth and grayscale difference of the catalyst, and surface chemical property parameters such as functional group peak height ratio. Then, a dataset partitioning module divides the sample dataset into a training set and a test set. A model training module uses the training set to train multiple candidate models, and a model selection module selects a prediction model from the candidate models. Finally, the prediction module inputs the catalyst data to be tested into the prediction model to obtain the predicted degradation rate output by the prediction model. The system can train candidate models using Bayesian optimization and select a prediction model by combining the evaluation indicators calculated from the test set. This allows for the use of multi-model fusion data to enhance the machine learning process in environments with small sample machine learning data volumes, thereby improving the prediction accuracy and generalization ability of catalyst activity analysis.

[0153] It should be noted that other corresponding descriptions of the functional units involved in the catalyst activity analysis system based on material surface physicochemical indicators provided in the embodiments of this application can be found in the corresponding descriptions in the catalyst activity analysis method based on material surface physicochemical indicators provided in the above embodiments, and will not be repeated here.

[0154] This application also provides a computer device, specifically a personal computer, server, network device, etc. The computer device includes a bus, processor, memory, and communication interface, and may also include input / output interfaces and a display device. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores location information. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0155] Those skilled in the art will understand that the structure of the computer device described above is only a partial structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components, or combine certain components, or have different component arrangements.

[0156] In one embodiment, a computer-readable storage medium is also provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0157] In one embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0158] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0159] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the methods described above.

[0160] Any references to memory, database, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0161] Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take many forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0162] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc.

[0163] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0164] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for analyzing catalyst activity based on physical and chemical indicators of a material surface, characterized by, The method comprises: obtaining a sample data set, the sample data set being historical experimental data with a label of a pollutant degradation rate; the sample data set comprising surface physical property parameters, surface chemical property parameters and catalytic reaction process condition parameters of a catalyst; the surface physical property parameters comprising a pore depth gray scale difference extracted from a scanning electron microscope image; the surface chemical property parameters comprising a functional group peak height ratio extracted from an infrared spectrum; the pore depth gray scale difference being a gray scale difference value of an orifice area and a hole bottom area determined along a central axis of a hole area in a gray scale image of a scanning electron microscope image of the catalyst; dividing the sample data set into a training set and a test set; training multiple candidate models using the training set, the candidate models being machine learning models obtained by searching for optimal hyperparameter combinations using a Bayesian optimization algorithm with the pollutant degradation rate as the prediction target; the multiple candidate models at least comprising a random forest model, an extreme gradient boosting model and a stacked ensemble model; selecting a prediction model from the candidate models, the prediction model being an optimal candidate model determined based on evaluation indexes by performing performance evaluation on the trained multiple candidate models using the test set; the evaluation indexes comprising at least one of a coefficient of determination, a mean absolute error and a root mean square error; obtaining to-be-tested catalyst data and inputting the to-be-tested catalyst data into the prediction model to obtain a predicted degradation rate output by the prediction model.

2. The method of claim 1, wherein, Obtaining a sample data set comprises: obtaining application information of a to-be-tested catalyst; constructing a keyword combination according to the application information, the keyword combination comprising a catalyst keyword, a pollutant keyword and a type keyword; retrieving real experimental data within a preset analysis period from a database based on the keyword combination; identifying the surface physical property parameters, the surface chemical property parameters and the catalytic reaction process condition parameters from the real experimental data to obtain the historical experimental data; labeling the historical experimental data with the pollutant degradation rate in the real experimental data to generate the sample data set.

3. The method of claim 2, wherein, Identifying the surface physical property parameters, the surface chemical property parameters and the catalytic reaction process condition parameters from the real experimental data to obtain the historical experimental data comprises: extracting a scanning electron microscope image of a catalyst from the real experimental data; converting the scanning electron microscope image into a gray scale image and identifying a hole area in the gray scale image; determining an orifice area and a hole bottom area along a central axis of the hole area, the orifice area being an annular area extending inward by a preset number of pixels from an edge of the hole area; the hole bottom area being a square area containing a deepest part of the hole area; calculating gray scale difference values of the orifice area and the hole bottom area respectively using an image histogram method to obtain the pore depth gray scale difference.

4. The method of claim 2, wherein, Identifying the surface physical property parameters, the surface chemical property parameters and the catalytic reaction process condition parameters from the real experimental data to obtain the historical experimental data comprises: extracting infrared spectrum of the catalyst from the real experimental data; reading stretching vibration peak height in the infrared spectrum, the stretching vibration peak height including carbonyl stretching vibration peak height, single bond carbon-oxygen bond stretching vibration peak height and iron-oxygen bond stretching vibration peak height; calculating functional group peak height ratio according to the stretching vibration peak height, the functional group peak height ratio including first peak height ratio and second peak height ratio; the first peak height ratio is the ratio of the carbonyl stretching vibration peak height to the single bond carbon-oxygen bond stretching vibration peak height; the second peak height ratio is the ratio of the iron-oxygen bond stretching vibration peak height to the single bond carbon-oxygen bond stretching vibration peak height.

5. The method of claim 1, wherein, dividing the sample data set into a training set and a test set, including: adopting an efficient machine learning regression algorithm based on a gradient boosting framework to fill in missing values in the sample data set by using a mean method; using Yeo-Johnson transformation to perform data transformation on the sample data set to generate a normally distributed data set; using a standardization scaler to standardize the normally distributed data set to obtain a standardized data set; dividing the standardized data set into a training set and a test set according to a five-fold cross-validation mechanism.

6. The method of claim 1, wherein, training multiple candidate models using the training set, including: inputting the training set into multiple candidate models, the model architectures of the multiple candidate models being all or partially different; obtaining training prediction results output by the candidate models; calculating training loss according to the training prediction results and pollutant degradation rate labels in the training set; performing iterative training based on the training loss to adjust model parameters of the candidate models.

7. The method of claim 6, wherein, performing iterative training based on the training loss to adjust model parameters of the candidate models, including: initializing iteration parameters, the iteration parameters including exploration times and optimization iteration times; performing iterative training on the candidate models according to the iteration parameters and the training loss; wherein the candidate models are configured to search for an optimal hyperparameter combination of the candidate models by using a Bayesian optimization algorithm to search for a hyperparameter space during the iterative training process; outputting model parameters of the candidate models according to the optimal hyperparameter combination.

8. The method of claim 7, wherein, performing iterative training on the candidate models according to the iteration parameters and the training loss, including: using a Bayesian optimization algorithm to search for an optimal hyperparameter combination of the random forest model and the extreme gradient boosting model respectively by taking average performance under a five-fold cross-validation strategy as an evaluation standard; generating meta-feature vectors of the random forest model and the extreme gradient boosting model based on the optimal hyperparameter combination; constructing a training data set by taking the meta-feature vectors as new input features and taking pollutant degradation rates in the sample data set as output labels; constructing a meta-learner using a least absolute shrinkage and selection operator cross-validation algorithm and performing model training on the training data set using the meta-learner to obtain the stacked ensemble model.

9. The method of claim 1, wherein, selecting a prediction model from the candidate models, including: inputting the test set into the trained multiple candidate models; obtaining test prediction results output by the candidate models; According to the test prediction result and the pollutant degradation rate label in the test set, an evaluation index is calculated; Based on the determination coefficient, a prediction accuracy evaluation is performed on the alternative models to determine a high-precision model with a determination coefficient higher than a preset coefficient threshold; The mean absolute error and the root mean square error are used to verify the high-precision model to determine an optimal alternative model.

10. A catalyst activity analysis system based on a physical-chemical index of a material surface, characterized by, The system comprises: A sample data acquisition module is configured to acquire a sample data set, which is historical experimental data with a pollutant degradation rate label; the sample data set includes surface physical property parameters, surface chemical property parameters, and catalytic reaction process condition parameters of a catalyst; the surface physical property parameters include a pore depth gray difference extracted from a scanning electron microscope image; the surface chemical property parameters include a functional group peak height ratio extracted from an infrared spectrum image; the pore depth gray difference is a gray difference value between a pore opening area and a pore bottom area determined along a central axis of the pore area in a gray image of a scanning electron microscope image of the catalyst; A data set division module is configured to divide the sample data set into a training set and a test set; A model training module is configured to train a plurality of alternative models using the training set, wherein the alternative model is a machine learning model obtained by searching for an optimal hyperparameter combination using a Bayesian optimization algorithm with a pollutant degradation rate as a prediction target; the plurality of alternative models at least include a random forest model, an extreme gradient boosting model, and a stacked ensemble model; A model selection module is configured to select a prediction model from the alternative models, wherein the prediction model is an optimal alternative model determined based on an evaluation index by performing performance evaluation on the trained plurality of alternative models using the test set; the evaluation index includes at least one of a determination coefficient, a mean absolute error, and a root mean square error; A prediction module is configured to acquire a to-be-tested catalyst data and input the to-be-tested catalyst data into the prediction model to obtain a predicted degradation rate output by the prediction model.

Images