Method of machine learning optimization of metal modified ets-10 catalyzed co2 methanation conditions

Abstract

The application discloses a method for optimizing metal modified ETS-10 catalytic CO2 methanation conditions by machine learning, which comprises the following steps: firstly, collecting CO2 methanation reaction experimental data of metal modified ETS-10 series catalysts, constructing a data set, and calculating CH4 yield; then, processing the data set, and establishing a CO2 methanation reaction performance prediction model based on an XGBoost regression model; setting a candidate reaction condition space of the metal modified ETS-10 series catalysts according to an existing experimental condition range, predicting and sorting CO2 conversion rate, CH4 selectivity and CH4 yield under the candidate conditions by using the trained model, and recommending the next round of experimental conditions; finally, adding the real experimental results of the recommended experimental conditions to the historical sample set, updating the prediction model, and obtaining relatively optimal catalyst composition and reaction conditions after multiple rounds of prediction, recommendation, experimental verification and model updating. The application can reduce the number of blind experiments, and improve the screening efficiency of CO2 methanation reaction conditions of the metal modified ETS-10 series catalysts.

Description

Technical Field

[0001] This invention belongs to the technical fields of CO2 catalytic conversion, methanation reaction condition optimization, and artificial intelligence-assisted catalyst screening, specifically involving a method for optimizing the conditions of metal-modified ETS-10 catalytic CO2 methanation using machine learning. Background Technology

[0002] In recent years, the massive consumption of fossil fuels has led to a continuous increase in greenhouse gas emissions such as CO2. The resource-based conversion of CO2 has become an important research direction in the fields of energy, chemical engineering, and carbon emission reduction. The CO2 methanation reaction can convert CO2 and H2 into CH4 and H2O under the action of a catalyst. CH4 can be used as a major component of natural gas, therefore this reaction has significant application value in CO2 utilization, hydrogen storage, and green energy conversion.

[0003] ETS-10 molecular sieves possess a unique framework structure, good thermal stability, and certain acid-base properties, making them suitable as carriers for metal active components in CO2 methanation reactions. By introducing metals or composite metal components such as Ni, Co, Cu, Zn, Fe, and Zr, the metal dispersion state, CO2 adsorption and activation capacity, and the conversion process of reaction intermediates can be controlled, thereby affecting CO2 conversion rate, CH4 selectivity, and CH4 yield.

[0004] However, the CO2 methanation performance of the metal-modified ETS-10 series catalysts is influenced by a variety of factors, including support type, active metal component, reaction temperature, reaction pressure, gas hourly space velocity, and catalyst dosage. Traditional single-factor experimental screening methods typically involve long experimental cycles, high gas and reagent consumption, and are difficult to quickly obtain the optimal reaction region under coupled multi-factor conditions. Therefore, it is necessary to establish an intelligent optimization method that can combine existing literature data with initial experimental data.

[0005] The XGBoost machine learning model can learn the nonlinear relationship between catalyst composition, reaction conditions, and catalytic performance from experimental data, and predict the CO2 methanation performance under untested conditions. Applying the XGBoost model to optimize CO2 methanation reaction conditions for metal-modified ETS-10 series catalysts can reduce the number of blind experiments, improve the efficiency of catalyst composition screening and reaction condition optimization, and provide an effective method for obtaining catalytic systems with high CO2 conversion and high CH4 selectivity. Summary of the Invention

[0006] To address the problems of numerous experiments, complex variable relationships, and low optimization efficiency in the screening of CO2 methanation reaction conditions for existing metal-modified ETS-10 series catalysts, this invention proposes a machine learning-based method for optimizing CO2 methanation reaction conditions for metal-modified ETS-10 series catalysts. This method integrates existing experimental data on CO2 methanation using metal-modified ETS-10 catalysts, establishes predictive relationships between catalyst support type, active metal component, reaction temperature, reaction pressure, gas hourly space velocity, catalyst dosage, and CO2 methanation performance, and achieves synergistic optimization of catalyst composition and reaction conditions through a closed-loop process of candidate condition prediction, experimental condition recommendation, experimental verification, and model updating.

[0007] The technical solution of the present invention is as follows: A method for optimizing CO2 methanation conditions catalyzed by metal-modified ETS-10 using machine learning, characterized by comprising the following steps: 1) Collect experimental data on the CO2 methanation reaction of metal-modified ETS-10 series catalysts to construct an initial sample set. This initial sample set includes catalyst support type, active metal component, reaction temperature, reaction pressure, gas hourly space velocity (GHSV), catalyst dosage, CO2 conversion rate, and CH4 selectivity. The catalyst support type includes one or both of ETS-10 and ETS-10(C). The active metal component includes one or more of Ni, Co, Cu, Fe, Zn, Zr, NiCo, NiCu, NiZn, NiFe, and NiZr.

[0008] 2) Calculate the CH4 yield based on the CO2 conversion rate and CH4 selectivity to obtain reaction performance data including CO2 conversion rate, CH4 selectivity and CH4 yield; The CH4 yield is calculated according to the following formula: CH4 yield = CO2 conversion × CH4 selectivity / 100.

[0009] 3) Perform data preprocessing on the initial sample set, including missing value handling, duplicate value handling, categorical variable encoding, and numerical variable handling, to obtain a machine learning training dataset; 4) Using catalyst support type, active metal component, reaction temperature, reaction pressure, gas hourly space velocity (GHSV), and catalyst dosage as input variables, and at least one of CO2 conversion, CH4 selectivity, or CH4 yield as output variables, an XGBoost regression prediction model is constructed; the reaction temperature is 280–400℃, the reaction pressure is 0.1–3.0 MPa, and the GHSV is 3600–15000 mL·gcat. - ¹·h - ¹, the amount of catalyst used is 0.1–4.0 g.

[0010] The XGBoost regression prediction model fits the nonlinear relationship between input and output variables through gradient boosting with multiple regression trees, and adjusts the model's prediction performance by adjusting the number of regression trees, maximum tree depth, learning rate, subsampling ratio, column sampling ratio, and regularization parameters.

[0011] 5) Based on the existing experimental conditions and the experimental conditions to be screened, the candidate reaction condition space of the metal-modified ETS-10 series catalysts is set. The candidate reaction condition space includes combinations of different catalyst support types, different metal active components, different reaction temperatures, different reaction pressures, different gas space velocities, and different catalyst dosages. 6) Use the trained XGBoost regression prediction model to predict the CO2 methanation performance of each group of candidate conditions in the candidate reaction condition space, and rank the candidate conditions according to the predicted CO2 conversion rate, CH4 selectivity or CH4 yield to obtain recommended experimental conditions. 7) Conduct CO2 methanation experiments to verify the results according to the recommended experimental conditions: gas velocity: 9600 mL·gcat -1 ·h -1 A CO2 methanation experiment was conducted under the following conditions: pressure: 3 MPa; temperature: 350℃; time: 9 h. Actual experimental feedback results were obtained. The recommended experimental conditions were selected from the top candidate conditions, ranked from highest to lowest based on predicted CO2 conversion rate, predicted CH4 selectivity, or predicted CH4 yield. The actual experimental feedback results were obtained and added to the historical sample set. The XGBoost regression prediction model was then updated. This process of prediction, recommendation, experimental verification, and model updating was repeated until the preset number of experimental rounds was reached or the target performance requirements were met, outputting the relatively optimal catalyst composition and reaction conditions. The XGBoost regression prediction model update process included adding the actual CO2 conversion rate, CH4 selectivity, and CH4 yield corresponding to the recommended experimental conditions to the historical sample set and retraining the XGBoost regression prediction model.

[0012] Preferably, the above method further includes using the SHAP feature importance analysis method to interpret the trained XGBoost regression prediction model, and to obtain the contribution of catalyst support type, metal active component, reaction temperature, reaction pressure, gas space velocity and catalyst dosage to the prediction results of CO2 methanation performance.

[0013] The beneficial effects of this invention are as follows: This invention introduces the XGBoost machine learning model into the optimization process of CO2 methanation reaction conditions of metal-modified ETS-10 series catalysts, which can predict the catalytic performance under different catalyst compositions and reaction conditions based on existing experimental data, reduce the number of blind experiments, and reduce experimental time and gas consumption costs; at the same time, this invention can simultaneously consider multiple factors such as catalyst support type, metal active component, reaction temperature, reaction pressure, gas space velocity and catalyst dosage, and is suitable for catalyst composition screening and reaction condition optimization under multivariate coupling conditions. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method for optimizing CO2 methanation reaction conditions of metal-modified ETS-10 series catalysts based on the XGBoost machine learning model of this invention.

[0015] Figure 2 This is a schematic diagram illustrating the construction of the CO2 methanation reaction dataset and its input and output variables in this invention.

[0016] Figure 3 This is a graph showing the cross-validation prediction results of the XGBoost regression model for CH4 yield in this invention.

[0017] Figure 4 This is a SHAP feature importance analysis diagram of the CH4 yield prediction results of different metal active components in this invention. Detailed Implementation

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings: like Figure 1 As shown, a method for optimizing the conditions of metal-modified ETS-10 catalytic CO2 methanation using machine learning includes the following steps: 1) In specific implementation, an initial sample set was constructed using 105 sets of experimental data obtained by the inventor through independent CO2 methanation experiments. For example... Figure 2As shown, this embodiment organizes the data obtained from CO2 methanation experiments into a dataset of reaction performance of metal-modified ETS-10 series catalysts. The dataset uses each catalyst composition and reaction condition as a sample. The catalyst support type (Support) and the active metal component (Metal) are used as catalyst composition variables, while the reaction temperature (Temperature), reaction pressure (Pressure), gas space velocity (GHSV), and catalyst dosage (Catalyst_mass) are used as reaction condition variables. These variables collectively constitute the input variables of the XGBoost regression prediction model. CO2 conversion and CH4 selectivity are obtained from CO2 methanation experiments, and the CH4 yield is calculated based on the CO2 conversion and CH4 selectivity, and is used as one of the model output variables.

[0019] Furthermore, on Figure 2 Missing values, duplicate values, and outliers in the dataset were checked and processed. Categorical variables such as Support and Metal were encoded. Numerical variables such as reaction temperature, reaction pressure, gas hourly space velocity, and catalyst dosage were input into the model as continuous variables, ultimately obtaining effective sample data that can be used for machine learning training and validation. Through this dataset construction method, the original CO2 methanation experimental data was transformed into a machine learning input feature matrix and output performance labels, providing a data foundation for subsequent XGBoost regression model training, cross-validation, and candidate reaction condition prediction.

[0020] The experimental data include catalyst support type, active metal component, reaction temperature, reaction pressure, gas hourly space velocity, catalyst dosage, CO2 conversion rate, and CH4 selectivity. The catalyst support types include ETS-10 and ETS-10(C), and the active metal components include Ni, Co, Cu, Fe, Zn, Zr, NiCo, NiCu, NiZn, NiFe, and NiZr.

[0021] Method 1: Measure 9.2 mL of water glass, then weigh 1.4 g of NaOH, dissolve it in 8 mL of water, cool the solution, and add it to the water glass. Stir for 0.5 h. Measure 5.9 mL of TiCl3 hydrochloric acid solution and add it dropwise to the mixed solution, stirring for 1 h. Weigh 1.76 g of KF, dissolve it in 5 mL of water, and add it dropwise to the above mixed system, stirring for 1.5 h. At 230 °C... o C The reaction was carried out for 72 hours. After the reaction was complete, the mixture was filtered, and 100 ml of solid was obtained. o After drying at C for 12 hours, grind the powder and then heat it at 450°C. o The sample was obtained by high-temperature calcination at C for 5 hours.

[0022] Method 2: Measure 9.2 mL of water glass, then weigh out 0.5 g of NaOH, dissolve it in 13 mL of water, cool it, and add it to the water glass. Stir for 0.5 h. Weigh out 2.81 g of potassium titanium oxalate dihydrate, add it to the mixed solution, and stir for 1 h. At 230 °C... o The reaction was carried out at C for 72 hours. After the reaction was completed, the mixture was filtered, the solid was dried at 100 °C for 12 hours, ground, and then subjected to a process at 450 °C. o The sample was obtained by high-temperature calcination at C for 5 hours.

[0023] Weigh 1g of ETS-10 zeolite molecular sieve sample, add deionized water dropwise, and measure the immersion rate of the sample, which is recorded as VmL.

[0024] (1) Weigh 0.3616 g Fe(NO3)3·9H2O, add V mL H2O, weigh 1 g ETS-10 zeolite molecular sieve sample, and add ferric nitrate solution dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace at 450 °C. o C, calcined for 5 hours.

[0025] (2) Weigh 0.247 g Co(NO3)2·6H2O, add V mL H2O, weigh 1 g ETS-10 zeolite molecular sieve sample, and add cobalt nitrate solution dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace at 450 °C. o C, calcined for 5 hours.

[0026] (3) Weigh 0.2478g Ni(NO3)2·6H2O, add V mL H2O, weigh 1g ETS-10 zeolite molecular sieve sample, and add nickel nitrate solution dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace at 450°C. o C, calcined for 5 hours.

[0027] (4) Weigh 0.19 g Cu(NO3)2·3H2O, add V mL H2O, weigh 1 g ETS-10 zeolite molecular sieve sample, and add copper nitrate solution dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace and heat at 450 °C. o C, calcined for 5 hours.

[0028] (5) Weigh 0.2275g Zn(NO3)2·6H2O, add V mL H2O, weigh 1g ETS-10 zeolite molecular sieve sample, and add zinc nitrate solution dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace and heat at 450°C. o C, calcined for 5 hours.

[0029] (6) Weigh 0.2354 g of Zr(NO3)4·5H2O, add V mL of H2O, weigh 1 g of ETS-10 zeolite molecular sieve sample, and add zirconium nitrate solution dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace and heat at 450 °C. o C, calcined for 5 hours.

[0030] (7) Weigh 0.124 g Ni(NO3)2·6H2O and 0.1808 g Fe(NO3)3·9H2O, add V mL H2O, weigh 1 g ETS-10 zeolite molecular sieve sample, and add nickel nitrate and ferric nitrate solutions dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace at 450 °C. o C, calcined for 5 hours.

[0031] (8) Weigh 0.1239 g Ni(NO3)2·6H2O and 0.1808 g Co(NO3)3·6H2O, add V mL H2O, weigh 1 g ETS-10 zeolite molecular sieve sample, and add nickel nitrate and ferric nitrate solutions dropwise to the sample powder and mix thoroughly. Place the final product in a muffle furnace and calcine at 450 °C for 5 h.

[0032] (9) Weigh 0.1239 g Ni(NO3)2·6H2O and 0.095 g Cu(NO3)2·3H2O, add V mL H2O, weigh 1 g ETS-10 zeolite molecular sieve sample, and add nickel nitrate and ferric nitrate solutions dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace at 450 °C. o C, calcined for 5 hours.

[0033] (10) Weigh 0.1239g Ni(NO3)2·6H2O and 0.11375g Zn(NO3)2·6H2O, add V mL H2O, weigh 1g ETS-10 zeolite molecular sieve sample, and add nickel nitrate and ferric nitrate solutions dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace at 450°C. o C, calcined for 5 hours.

[0034] (11) Weigh 0.1239 g Ni(NO3)2·6H2O and 0.1177 g Zr(NO3)4·5H2O, add V mL of H2O, weigh 1 g of ETS-10 zeolite molecular sieve sample, and add nickel nitrate and ferric nitrate solutions dropwise to the sample powder, mixing thoroughly. Place the final product in a muffle furnace at 450 °C. o C, calcined for 5 hours.

[0035] 2) Data preprocessing of the initial sample set. Specifically, the support type and active metal component are coded as categorical variables, while reaction temperature, reaction pressure, gas hourly space velocity, and catalyst dosage are input into the model as numerical variables. CO2 conversion and CH4 selectivity are used as reaction performance data, and the CH4 yield is calculated according to the following formula:

[0036] Among them, Y CH4 Xco2 represents the CH4 yield, Xco2 represents the CO2 conversion, and S represents the CO2 conversion. CH4 CH4 selectivity is represented by the expression "CH4 selectivity". In the specific dataset, support type, active metal component, reaction temperature, reaction pressure, gas hourly space velocity (GHSV), and catalyst dosage correspond to Support, Metal, Temperature (C), Pressure (MPa), GHSV (mL gcat h), and Catalyst mass (g), respectively; CO2 conversion, CH4 selectivity, and CH4 yield correspond to CO2 conversion pct, SCH4 pct, and CH4 yield pct, respectively. The processed sample data was stored in a CSV file, and the pandas library in Python was used for data reading, missing value handling, duplicate value handling, and variable encoding, ultimately obtaining 96 sets of valid sample data.

[0037] 3) Establish an XGBoost regression prediction model. Specifically, Support, Metal, Temperature C, Pressure MPa, GHSV mL gcat h, and Catalyst mass g are used as input variables, and CH4 yield pct is used as the output variable to train the XGBoost regression model. During model training, the number of regression trees is set to 100, the maximum tree depth to 2, the learning rate to 0.03, the subsampling ratio to 0.8, the column sampling ratio to 0.8, the minimum child node weight to 2, the L1 regularization parameter to 0.1, the L2 regularization parameter to 2.0, and the random seed to 42. The number of regression trees, maximum tree depth, and learning rate primarily affect the model's fitting ability and complexity; the subsampling ratio and column sampling ratio are used to reduce the model's over-reliance on certain samples or features, reducing the risk of overfitting; the minimum child node weight is used to limit the further splitting of nodes with small sample sizes; the L1 and L2 regularization parameters are used to constrain model complexity and improve the model's generalization ability; and the random seed is used to ensure the reproducibility of data partitioning and model training results. Unless otherwise specified, the default values ​​of the XGBoost model are used. Their values ​​may have some impact on training speed and prediction results, but within the data scale and parameter range of this embodiment, they will not change the overall optimization method and technical effect of the model.

[0038] 4) To evaluate the model's predictive performance, the sample data was divided into training and test sets at a ratio of 80% and 20%, respectively, and the model was further validated using a 5-fold cross-validation method. The cross-validation results showed that the model's prediction of CH4 yield was CV R² = 0.835, CV RMSE = 7.186, and the experimental and predicted values ​​were generally distributed near the diagonal of y=x. Figure 3 As shown, the model predictions and experimental values ​​generally have good consistency, indicating that the established XGBoost model can be used to predict the CO2 methanation performance of metal-modified ETS-10 series catalysts and recommend experimental conditions.

[0039] 5) Constructing a candidate reaction condition space. In practice, based on the existing experimental range and actual reaction apparatus conditions, combinations of different ETS-10 support types, different metal active components, different reaction temperatures, different reaction pressures, different gas space velocities, and different catalyst dosages are set to form a set of candidate experimental conditions. To avoid over-extrapolation of the model, candidate conditions are preferentially set near the range of existing experimental data.

[0040] In practice, the XGBoost regression model establishes a nonlinear mapping relationship between input and output variables through an ensemble of multiple regression trees. Let the experimental conditions of the i-th group be x. i Its corresponding CO2 methanation performance is y i Then the model prediction is expressed as:

[0041] Where, x i The input variables, y, consist of the ETS-10 support type, the active metal component, the reaction temperature, the reaction pressure, the gas hourly space velocity, and the catalyst dosage. i F(x) represents the yield of CH4. i ) represents the prediction function built by the XGBoost regression model.

[0042] The XGBoost regression model is obtained by summing multiple regression trees, and its prediction function can be expressed as:

[0043] Here, fk(x) represents the k-th regression tree, and K represents the number of regression trees. The model gradually improves its predictive ability for CO2 methanation performance by continuously fitting the prediction error of the previous round. During training, the risk of overfitting is reduced and the model's predictive ability for unexperimented conditions is improved by adjusting the number of regression trees, maximum tree depth, learning rate, subsampling ratio, and regularization parameters.

[0044] In the candidate condition recommendation process, the candidate response condition space is denoted as Xcandidate. The trained XGBoost model is used to predict each set of response conditions in the candidate space to obtain the corresponding prediction performance value:

[0045] Then, the predicted values ​​ŷ are sorted from highest to lowest, and several sets of conditions with higher predicted CH4 yields are selected as the recommended experimental conditions for the next round. The optimal candidate conditions can be expressed as:

[0046] in, This represents the relatively optimal reaction conditions recommended by the model.

[0047] The experimenters followed the recommended conditions Conduct CO2 methanation experiments to obtain actual experimental feedback values. The new data set is then added to the historical sample set, and the updated sample set is represented as follows:

[0048] Among them, D old D represents the historical sample set before the update. new This represents the sample set after incorporating new experimental feedback. The XGBoost regression model is then retrained, and the process of "candidate condition prediction—recommendation experiment—experimental feedback—model update" is repeated until the preset number of experimental rounds is reached or the catalyst composition and reaction conditions that meet the target performance requirements are obtained.

[0049] 6) Conduct CO2 methanation experiments according to the recommended experimental conditions to obtain the corresponding CO2 conversion rate, CH4 selectivity, and CH4 yield. Add the newly obtained experimental feedback results to the historical sample set and retrain the XGBoost regression model. Through multiple rounds of "model prediction - experimental verification - data update - model retraining" cycle, gradually improve the model's predictive ability.

[0050] 7) Furthermore, the SHAP method can be used to analyze the contribution of metallic features to the trained XGBoost model. For example... Figure 4 As shown, SHAP analysis can reflect the contribution of different active metal components to the predicted CH4 yield. This analysis can help determine the influence of different metal components on the CO2 methanation performance, providing a reference for subsequent catalyst composition screening and reaction condition optimization.

[0051] 8) This invention utilizes the XGBoost machine learning model to transform existing literature data and initial experimental data into a reaction condition optimization model that can be used for prediction and screening. It eliminates the need for numerous single-factor experiments, enabling rapid recommendations for CO2 methanation reaction conditions of the metal-modified ETS-10 series catalysts. Compared to traditional empirical screening methods, this invention reduces the number of experiments, lowers experimental costs, and improves the efficiency of catalyst composition screening and reaction condition optimization.

Claims

1. A method for optimizing CO2 methanation conditions catalyzed by metal-modified ETS-10 using machine learning, characterized in that, Includes the following steps: 1) Collect experimental data on CO2 methanation reaction of metal-modified ETS-10 series catalysts and construct an initial sample set, which includes catalyst support type, metal active component, reaction temperature, reaction pressure, gas space velocity, catalyst dosage, CO2 conversion rate and CH4 selectivity; 2) Calculate the CH4 yield based on the CO2 conversion rate and CH4 selectivity to obtain reaction performance data including CO2 conversion rate, CH4 selectivity and CH4 yield; 3) Perform data preprocessing on the initial sample set, including missing value handling, duplicate value handling, categorical variable encoding, and numerical variable handling, to obtain a machine learning training dataset; 4) Using catalyst support type, active metal component, reaction temperature, reaction pressure, gas space velocity and catalyst dosage as input variables, and at least one of CO2 conversion rate, CH4 selectivity or CH4 yield as output variables, construct an XGBoost regression prediction model. 5) Based on the existing experimental conditions and the experimental conditions to be screened, the candidate reaction condition space of the metal-modified ETS-10 series catalysts is set. The candidate reaction condition space includes combinations of different catalyst support types, different metal active components, different reaction temperatures, different reaction pressures, different gas space velocities, and different catalyst dosages. 6) Use the trained XGBoost regression prediction model to predict the CO2 methanation performance of each group of candidate conditions in the candidate reaction condition space, and rank the candidate conditions according to the predicted CO2 conversion rate, CH4 selectivity or CH4 yield to obtain recommended experimental conditions. 7) Conduct CO2 methanation experiments to verify the results according to the recommended experimental conditions, obtain real experimental feedback results, add the real experimental feedback results to the historical sample set, update the XGBoost regression prediction model, repeat the prediction, recommendation, experimental verification and model update process until the preset number of experimental rounds is reached or the target performance requirements are met, and output the relatively optimal catalyst composition and reaction conditions.

2. The method for optimizing the conditions for CO2 methanation catalyzed by metal-modified ETS-10 using machine learning according to claim 1, characterized in that, The catalyst support type includes one or both of ETS-10 and ETS-10(C).

3. The method for optimizing the conditions for CO2 methanation catalyzed by metal-modified ETS-10 using machine learning according to claim 1, characterized in that, The active metal component includes one or more of Ni, Co, Cu, Fe, Zn, Zr, NiCo, NiCu, NiZn, NiFe, and NiZr.

4. The method for optimizing the conditions for CO2 methanation catalyzed by metal-modified ETS-10 using machine learning according to claim 1, characterized in that, The reaction temperature is 280–400℃, the reaction pressure is 0.1–3.0 MPa, and the gas hourly space velocity is 3600–15000 mL·gcat. - ¹·h - ¹, the amount of catalyst used is 0.1–4.0 g.

5. The method according to claim 1, characterized in that, The CH4 yield is calculated according to the following formula: CH4 yield = CO2 conversion × CH4 selectivity / 100.

6. The method for optimizing the conditions for CO2 methanation catalyzed by metal-modified ETS-10 using machine learning according to claim 1, characterized in that, The XGBoost regression prediction model fits the nonlinear relationship between input and output variables through gradient boosting with multiple regression trees, and adjusts the model's prediction performance by adjusting the number of regression trees, maximum tree depth, learning rate, subsampling ratio, column sampling ratio, and regularization parameters.

7. The method for optimizing the conditions for CO2 methanation catalyzed by metal-modified ETS-10 using machine learning according to claim 1, characterized in that, The recommended experimental conditions for conducting CO2 methanation experiments are the top few candidate conditions selected after ranking them from high to low based on predicted CO2 conversion rate, predicted CH4 selectivity, or predicted CH4 yield.

8. The method for optimizing the conditions for CO2 methanation catalyzed by metal-modified ETS-10 using machine learning according to claim 1, characterized in that, The XGBoost regression prediction model update process includes adding the actual CO2 conversion rate, CH4 selectivity, and CH4 yield corresponding to the recommended experimental conditions to the historical sample set, and retraining the XGBoost regression prediction model.

9. The method for optimizing the conditions for CO2 methanation catalyzed by metal-modified ETS-10 using machine learning according to claim 1, characterized in that, The method also includes using SHAP feature importance analysis to interpret the trained XGBoost regression prediction model, and obtaining the contribution of catalyst support type, active metal component, reaction temperature, reaction pressure, gas space velocity and catalyst dosage to the prediction results of CO2 methanation performance.

Images