Supergravity enhanced decarburization mass transfer performance prediction method based on machine learning

Abstract

The invention relates to the technical field of supergravity enhanced decarburization, in particular to a supergravity enhanced decarburization mass transfer performance prediction method based on machine learning, which comprises the following steps of: preprocessing mass transfer performance experimental data in a supergravity enhanced decarburization process; screening main influence parameters of mass transfer performance in the supergravity enhanced decarburization process; establishing a machine learning model for predicting the mass transfer performance of the supergravity enhanced decarburization process; establishing a multivariate nonlinear regression model for predicting the mass transfer performance of the supergravity enhanced decarburization process; comparing and evaluating the prediction model of the mass transfer performance of the supergravity enhanced decarburization process; predicting the mass transfer performance of the supergravity enhanced decarburization process; in this way, the mass transfer performance of the supergravity enhanced decarburization process can be accurately predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supergravity intensified decarburization, and particularly relates to a supergravity intensified decarburization mass transfer performance prediction method based on machine learning. BACKGROUND

[0002] Carbon capture, utilization and storage (CCUS) technology is an important means to realize low-carbonization of fossil energy. As the first step in the CCUS process, the cost of carbon capture accounts for 60% to 85% of the entire project, so its economic benefit is crucial to the CCUS project. The supergravity reactor is a new type of intensified mass transfer equipment, whose working principle is to simulate the supergravity environment by rotating the filler driven by the rotor to form a strong centrifugal field. In the supergravity field of hundreds to thousands of times of the earth's gravity, the material flows in the filler, and the strong shear force decomposes the liquid material into micron or even nanometer liquid elements, and the rapidly updated phase interface significantly improves the mass transfer rate between gas and liquid, reaching 1-3 orders of magnitude of the traditional filler tower. Therefore, the supergravity reactor is expected to provide a new technical solution for carbon capture in CCUS.

[0003] However, since the mass transfer performance in the supergravity intensified carbon capture process is simultaneously affected by the reactor type, the filler structure parameters, the fluid physical property parameters, and the operation parameters and other factors, it is difficult to accurately predict the mass transfer performance in the supergravity intensified decarburization process, and the existing technical methods cannot meet the basic requirements of quickly and accurately predicting the supergravity mass transfer performance, resulting in the difficulty in large-scale promotion of the supergravity intensified carbon capture technology. Therefore, it is necessary to propose a prediction method capable of quickly and accurately predicting the mass transfer performance in the supergravity intensified carbon capture process. SUMMARY

[0004] The present application aims to provide a supergravity intensified decarburization mass transfer performance prediction method based on machine learning, which aims to solve the technical problems in the prior art that the mass transfer performance in the supergravity intensified decarburization process is simultaneously affected by the reactor type, the filler structure parameters, the fluid physical property parameters, and the operation parameters and other factors, resulting in the difficulty in quickly and accurately predicting the mass transfer performance in the supergravity intensified decarburization process, and the inability to effectively guide the supergravity structure design and performance optimization.

[0005] To achieve the above-mentioned purpose, the present application provides a supergravity intensified decarburization mass transfer performance prediction method based on machine learning, comprising the following steps: preprocessing the mass transfer performance experimental data in the supergravity intensified decarburization process; screening the main influencing parameters of the mass transfer performance in the supergravity intensified decarburization process; establishing a machine learning model for predicting the mass transfer performance in the supergravity intensified decarburization process; A multi-element nonlinear regression model for predicting the mass transfer performance of the supergravity intensified decarburization process is established. The prediction model for the mass transfer performance of the supergravity intensified decarburization process is evaluated by comparison. The mass transfer performance of the supergravity intensified decarburization process is predicted.

[0006] In the step of preprocessing the experimental data of the mass transfer performance in the supergravity intensified decarburization process, the preprocessing process is as follows: The variable parameters affecting the mass transfer performance of the supergravity intensified decarburization process are determined, mainly including equipment structure parameters, gas-liquid physical property parameters, and operating parameters. The structure parameters mainly include the inner radius, outer radius, average radius, hydraulic diameter, thickness, porosity, and specific surface area of the filler. The gas-liquid physical property parameters mainly include the dynamic viscosity of the gas-liquid phase and the diffusion coefficient of CO2 in the gas-liquid phase. The operating parameters mainly include the gas-liquid phase flow rate, gas-liquid phase superficial velocity, and filler rotational speed. The above influencing parameters are converted into dimensionless variables using dimensionless analysis method.

[0007] In the step of converting the above influencing parameters into dimensionless variables using dimensionless analysis method, the dimensionless analysis process is as follows: The gas Reynolds number is the ratio of gas viscous force to inertial force, the liquid Reynolds number is the ratio of liquid viscous force to inertial force, the liquid Schmidt number is the ratio of liquid viscosity coefficient to diffusion coefficient, the gas Grashof number is the ratio of centrifugal force to viscous force, and the liquid-gas molar ratio is the ratio of liquid molar mass to gas molar mass concentration. The Sherwood number is the ratio of mass transfer coefficient to gas diffusion coefficient and specific surface area of the filler.

[0008] In the step of screening the main influencing parameters of the mass transfer performance in the supergravity intensified decarburization process, the screening process is as follows: The correlation degree between the main influencing parameters and the mass transfer performance is evaluated using four multi-factor correlation statistical analysis methods. The correlation degree between the influencing parameters and the mass transfer performance is evaluated using gray correlation coefficient analysis method, Pearson coefficient analysis method, Kendall coefficient analysis method, and Spearman coefficient analysis method. The input variables are screened according to the correlation coefficient size, the input variables with low correlation degree are removed, and the number of latest input variables of the machine learning model is determined.

[0009] In the step of establishing the machine learning model for predicting the mass transfer performance of the supergravity intensified decarburization process, the establishment process is as follows: Data preparation and partitioning: The randperm function is used to randomly partition the collected raw dataset into a training set and a test set. The training set is used for model training and parameter optimization, while the test set is used for model performance verification. Specifically, the total samples are randomly allocated according to a certain proportion to ensure the representativeness of the data distribution. Data normalization is performed to eliminate the impact of differences in the scale of different features on the model. The maximum-minimum normalization method is used to map the input features and the output target variable to the [0,1] interval. This process includes: normalizing the training set data and recording the normalization parameters; the test set data is transformed using the same normalization parameters as the training set to maintain the consistency of data processing. Model initialization, selecting radial basis functions to handle mapping relationships, and providing the value ranges of regularization parameters and kernel function parameters; Hyperparameter optimization employs a grid search combined with k-fold cross-validation to automatically find the optimal hyperparameter combination and determine the optimal parameter values ​​by minimizing the mean square error of cross-validation. Model training involves using optimized hyperparameters to train the model based on the training set data, and solving for the model parameters α Lagrange multiplier vector (α) and bias term (b). Model prediction and inverse normalization: The trained model is used to predict the test set to obtain normalized prediction values; then, the inverse transformation of the normalized parameters of the training set is used to restore the prediction results to the original dimensional space to obtain the final prediction values.

[0010] In the step of optimizing hyperparameters using grid search combined with k-fold cross-validation, the optimization process is as follows: Calculate the mean and standard deviation for each feature; Evolutionary standardization is applied to each eigenvalue; A logarithmic scale is used to ensure full coverage of the parameter space; The system iterates through all parameter combinations; The dataset is split into five parts; Cross-validation was performed on each set of data, and the mean squared error and standard deviation were calculated. Record the performance metrics for each set of kernel parameters; Choose the optimal kernel parameters and train the model on the complete dataset using the optimal kernel parameters; Predict the mass transfer performance of the decarbonization process enhanced by supergravity.

[0011] In the step of establishing a multivariate nonlinear regression model to predict the mass transfer performance of the hypergravity-enhanced decarbonization process, the establishment process is as follows: Data preprocessing; Dataset classification; Establish a multivariate nonlinear regression model; The regression coefficients are obtained using the NLINFIT function; Predict the mass transfer performance of the decarbonization process enhanced by supergravity.

[0012] In the comparative evaluation of the mass transfer performance of the supergravity-enhanced decarbonization process, the comparative evaluation process is as follows: Two model parameters are selected for evaluation: coefficient of determination and root mean square error, to evaluate the accuracy of the machine learning model. The closer the coefficient of determination is to 1, the higher the model accuracy; the closer it is to 0, the lower the model accuracy. This is suitable for evaluating the overall predictive accuracy of the model. The closer the mean relative error, root mean square error, and standard deviation are to 0, the better the model's predictive performance. The machine learning model and the multivariate nonlinear regression model are compared and evaluated using training and test set data to further assess the accuracy of the machine learning model. If the prediction results of the machine learning model are better than those of the multivariate nonlinear regression model, the optimized machine learning model can be used to predict the mass transfer performance in the hypergravity-enhanced decarbonization process. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0014] Figure 1 This is a flowchart of the steps of the method for predicting the mass transfer performance of the supergravity-enhanced decarbonization process based on a machine learning model, as proposed in this invention.

[0015] Figure 2 This is a flowchart of the pre-processing mass transfer performance pre-experiment data steps of the present invention.

[0016] Figure 3 This is a flowchart of the steps for screening the main parameters affecting mass transfer performance in this invention.

[0017] Figure 4 This is a flowchart of the LSSVM model steps for predicting mass transfer performance according to the present invention.

[0018] Figure 5 This is a flowchart illustrating the steps of the present invention for optimizing hyperparameters using a grid search combined with k-fold cross-validation.

[0019] Figure 6 This is a flowchart of the steps of the multivariate nonlinear regression model of the present invention.

[0020] Figure 7 This invention uses a graph showing the magnitude of the correlation between input and output variables.

[0021] Figure 8This is a graph showing the predicted mass transfer performance of the optimized machine learning model for the hypergravity-enhanced decarbonization process according to the present invention.

[0022] Figure 9 This is a graph showing the predicted mass transfer performance of the supergravity-enhanced decarbonization process using the multivariate nonlinear regression model of this invention. Detailed Implementation

[0023] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0024] Please see Figures 1 to 6 This invention provides a machine learning-based method for predicting the performance of enhanced carbon transfer in supergravity, comprising the following steps: S1: Preprocessing of experimental data on mass transfer performance during hypergravity-enhanced decarbonization; S2: Screening of the main parameters affecting mass transfer performance during the hypergravity-enhanced decarbonization process; S3: Establish a machine learning model to predict the mass transfer performance of the hypergravity-enhanced decarbonization process; S4: Establish a multivariate nonlinear regression model to predict the mass transfer performance of the hypergravity-enhanced decarbonization process; S5: Comparative evaluation of predictive models for mass transfer performance in hypergravity-enhanced decarbonization processes; S6: Predict the mass transfer performance of the hypergravity-enhanced decarbonization process.

[0025] In this embodiment, the experimental data on mass transfer performance during the enhanced decarbonization process under hypergravity are first preprocessed; then, the main influencing parameters of mass transfer performance are screened; next, a machine learning model for predicting the mass transfer performance during the enhanced decarbonization process under hypergravity is established; then, a multivariate nonlinear regression model for predicting the mass transfer performance during the enhanced decarbonization process under hypergravity is established; then, the machine learning models for predicting the mass transfer performance during the enhanced decarbonization process under hypergravity are compared and evaluated; finally, mass transfer performance is predicted, thus achieving accurate prediction of the mass transfer performance during the enhanced decarbonization process under hypergravity.

[0026] Furthermore, in the preprocessing step of the mass transfer performance experimental data during the hypergravity-enhanced decarburization process: S11: Determine the variable parameters that affect the mass transfer performance of the hypergravity-enhanced decarbonization process; S12: Classified according to equipment structural parameters, gas-liquid physical property parameters, and operating parameters; S13: Use dimensionless analysis to transform the influencing parameters into different dimensionless criterion numbers.

[0027] In this embodiment, the parameters affecting mass transfer performance during the enhanced decarbonization process under high gravity are determined. These parameters are categorized into structural parameters, physical property parameters, and operational parameters. Structural parameters include the inner radius, outer radius, average radius, hydraulic diameter, thickness, porosity, and specific surface area of ​​the packing material. Physical property parameters include the dynamic viscosity of the gas and liquid phases and the diffusion coefficient of CO2 in the gas and liquid phases. Operational parameters include the gas and liquid phase flow rates, apparent velocities of the gas and liquid phases, and packing rotation speed. Dimensionless analysis is used to convert these parameters and mass transfer coefficients into dimensionless variables: gas Reynolds number (ReG), liquid Reynolds number (ReL), liquid Schmidt number (ScL), gas Grashov number (GrG), liquid-gas molar ratio (Mr), and Sherwood number (Sh). The expressions are: In the formula, and These are the gas and liquid phase velocities, respectively, in m / s; and , respectively, are the dynamic viscosities of the gas and liquid phases, in Pa·s; and These are the diffusion coefficients of CO2 in the gas and liquid phases, respectively. and Gas and liquid flow rates (m) are respectively 3 / s; and The values ​​are the molar concentrations of CO2 in the inlet gas phase and the NaOH solution, respectively, in mol / L. and Here, represent the inner and outer radii of the packing material, respectively, in meters (m). Let be the average radius of the packing material, in meters (m). Let be the hydraulic radius, in meters. The rotational speed of the packing material is given in rad / s. The specific surface area of ​​the packing material is given in m². -1 ; Let s be the overall gas-side volumetric mass transfer coefficient. -1 .

[0028] It is worth noting that, due to the different airflow directions, The expressions differ for countercurrent and cross-flow supergravity reactors.

[0029] For a counter-current hypergravity reactor, the expression is: In the formula, z is the height of the packing material, in meters (m).

[0030] For a cross-flow hypergravity reactor, the expression is: ReG, ReL, ScL, GrG, and Mr are used as input variables for the model, and Sh is used as the output variable.

[0031] The experimental data came from different researchers, who used different reactor structures and operating conditions, resulting in a wide range of data. In order to accurately train the machine learning model, a random method was used to group all the data.

[0032] Furthermore, in the step of screening the main parameters affecting mass transfer performance during the hypergravity-enhanced decarburization process: S21: Identify four methods for multi-factor association statistical analysis; S22: Calculate the correlation coefficients corresponding to different analysis methods; S23: Compare the correlation coefficients of different analysis methods and eliminate input variables with low correlation. S24: Optimize the input variables for the machine learning model.

[0033] In this implementation, since there are five input variables affecting mass transfer performance and numerous influencing parameters, to reduce the number of input variables and improve the prediction accuracy of the machine learning model, it is necessary to filter out and eliminate input variables with less impact on the output variables. First, four multi-factor correlation statistical analysis methods are determined: grey relational coefficient, Pearson correlation coefficient, Kendall correlation coefficient, and Spearman correlation coefficient. Using relevant data, the correlation coefficients between the input and output variables for each analysis method are calculated. The correlation coefficients between the input and output variables are compared across different analysis methods to comprehensively evaluate the degree of correlation between the input and output variables, eliminating variables with low correlation from the existing five input variables. Finally, the optimal input variables for the final machine learning model are selected. The calculation formulas for the four multi-factor correlation statistical analysis methods are as follows: Grey relational coefficient:

[0034] In the formula, The resolution coefficient (ρ) ranges from 0 to 1. A larger ρ value indicates a stronger correlation between the input and output parameters. This is because... Correlation coefficient It is a comparison SeriesThe degree of association values with the reference sequence, so there are more than one of them, and the information is too scattered to facilitate holistic comparison. Therefore, it is necessary to concentrate the correlation coefficients into one value, that is, to find their average value, as a quantitative representation of the degree of association between the comparison sequence and the reference sequence. Correlation degree The calculation formula is as follows: Pearson coefficient:

[0035] In the formula , , and are the k-th input, the average value of all inputs, the i-th output, and the average value of all outputs respectively.

[0036] The Spearman correlation coefficient, also known as the rank correlation coefficient, is a measure that measures the correlation based on the ranks of random variables rather than their original values. It can be obtained by calculating the method of the Pearson coefficient calculation method. This method only needs to replace the original data in the original random variable with its rank in the random variable. Its expression is as follows: The Kendall correlation coefficient, also known as the concordance coefficient, is a hierarchical correlation coefficient. Its calculation method is as follows: For two pairs of observations Xi, Yi and Xj, Yj of X and Y, if Xi < Yi and Xj < Yj, or Xi > Yi and Xj > Yj, then these two pairs of observations are called concordant, otherwise they are discordant.

[0037] Negative numbers indicate negative correlation, and positive numbers indicate positive correlation. Under the premise of significance, the larger the absolute value, the stronger the correlation.

[0038] Furthermore, in the steps of establishing a machine learning model for predicting the mass transfer performance of the supergravity enhanced decarbonization process: S31: Dataset preparation and division. Use the randperm function to randomly divide the collected original dataset into a training set and a test set, and randomly allocate the total samples according to a certain ratio (such as 70%:30%); S32: Data normalization processing. The maximum-minimum normalization method maps the input features and output target variables to the interval [0,1]; S33: Model initialization. Select the radial basis function to process the mapping relationship, and give the value ranges of the regularization parameter and the kernel function parameter; S34: Hyperparameter optimization. Adopt the method of grid search combined with k-fold cross-validation to automatically find the optimal hyperparameter combination, and determine the optimal parameter value by minimizing the cross-validation mean square error (MSE); S35: Model training, using optimized hyperparameters, train the model based on the training set data, and solve for the model parameters α (Lagrange multiplier vector) and b (bias term); S36: Model prediction and inverse normalization. The trained model is used to predict the test set to obtain normalized prediction values. Then, the inverse transformation of the normalized parameters of the training set is used to restore the prediction results to the original dimensional space to obtain the final prediction values.

[0039] In this implementation, data preparation and partitioning are performed first. The `randperm` function is used to randomly divide the collected raw dataset into a training set and a test set. The training set is used for model training and parameter optimization, while the test set is used for model performance validation. Specifically, the total samples are randomly allocated according to a certain ratio (e.g., 70%:30%) to ensure the representativeness of the data distribution. Since mass transfer performance involves more than a dozen influencing factors, and these factors usually have different dimensions, data normalization is required to eliminate the influence of dimensions, provide numerical computation stability, and accelerate parameter optimization. The min-max normalization method is used to normalize the experimental data samples. , , , , Normalization is performed on feature parameters such as Sh, and the maximum and minimum normalization formulas are as follows: The data is after normalization; This is the original data; It is the minimum value among all data; This represents the maximum value among all data. The above method maps all data to the interval [0,1].

[0040] For model initialization, radial basis functions are selected to handle mapping relationships. The expression for the radial basis kernel function is as follows:

[0041] In the formula σ is the bandwidth parameter; the larger σ is, the "wider" the kernel function, and the slower the decay of similarity between samples. The Euclidean distance between two sample points.

[0042] The regularization parameter γ is given to have a range of values ​​[0.1, 100], with an initial value of γ = 10; kernel parameters are also given. The value range is [0.01, 100], and the initial value is... =1.

[0043] Regularization parameters and kernel parameters are model structure parameters that directly determine model performance. The γ parameter controls the balance between model complexity and fitting accuracy; too small a value leads to underfitting, while too large a value leads to overfitting. The σ² parameter determines the effective range of the kernel function; too small a value causes the model to degenerate into a "lookup table," resulting in extremely poor generalization ability; too large a value causes the model to degenerate into a linear model, unable to capture nonlinear relationships. These two hyperparameters cannot be automatically obtained during training like weight parameters, nor do they have universal default values. Different datasets and different problems require different combinations of hyperparameters. The dimensionality of the data, sample size, noise level, and degree of nonlinearity all influence the optimal parameter selection. Practice shows that the performance difference between optimized and unoptimized hyperparameters can be 2-5 times or even more. Hyperparameter optimization is like adjusting the focal length and aperture of a camera: no adjustment may result in a blurry or overexposed photo; arbitrary adjustment leads to unstable results; precise adjustment yields a clear and perfect photo. A grid search combined with k-fold cross-validation method is used to automatically find the optimal hyperparameter combination.

[0044] Model training involves using optimized hyperparameters to train the model based on the training set data, and solving for the model parameters: Lagrange multiplier vector (α) and bias term (b), as detailed below: Calculate the kernel matrix Ω. For N samples in the training set, calculate the N×N kernel matrix. Construct coefficient matrix A Construct the right-hand vector B The model parameters are obtained by solving the linear equations using the conjugate gradient method. Model prediction and denormalization: The trained model is used to predict the test set to obtain normalized predicted values. At this time, the predicted values ​​are all dimensionless values ​​in the [0,1] interval, which cannot directly reflect the actual values. Denormalization is required to restore the prediction results to the original dimensional space and obtain the final predicted values, so that the prediction results have practical significance.

[0045] Furthermore, in the step of automatically finding the optimal hyperparameter combination using a grid search combined with k-fold cross-validation: S41: Calculate the mean and standard deviation for each feature; S42: Standardize each eigenvalue; S43: Use logarithmic scaling to ensure full coverage of the parameter space; S44: The system iterates through all parameter combinations; S45: Perform a 5-fold dataset split for each set of parameters, and perform the following operation for each fold: Build a machine learning model using the current parameters (γ, σ2); S46: Perform cross-validation on each set of data and calculate the mean MES and standard deviation; S47: Record the performance indicators of each set of kernel parameters; S48: Select the optimal kernel parameters and train the model on the complete dataset using the optimal kernel parameters; S49: Predict the mass transfer performance of the supergravity-enhanced decarbonization process.

[0046] In this implementation, the parameter search space is first defined. Since the influence of different parameters is often exponential, a logarithmic scale is used to partition the space, which can more evenly cover different magnitudes and avoid over-dense sampling in a certain range, ensuring sufficient coverage of the parameter space. The ranges of the two kernel parameters, γ and σ², are defined. Grid search is an exhaustive parameter optimization method. It systematically traverses all parameter combinations in the predefined parameter space. Then, 5-fold cross-validation is performed, dividing the dataset into 5 folds: Fold 1: Training set: 1~80% of samples, Validation set: 81%~100% of samples; Fold 2: Training set: 1~60% of samples + 81%~100% of samples, Validation set: 61%~80% of samples; Fold 3: Training set: 1~40% of samples + 61%~100% of samples, Validation set: 41%~60% of samples; Fold 4: Training set: 1~ The training set consists of 20% of the samples plus 41%~100% of the samples, and the validation set consists of 21%~40% of the samples. In the fifth fold, the training set consists of 21%~100% of the samples, and the validation set consists of 1%~20% of the samples. For each fold dataset, a machine learning model is constructed using the current hyperparameters. The kernel matrix K is calculated, an augmented linear equation system is constructed, b and α are solved, and predictions are made on the validation set. The mean squared error (MSE) of fold 1 (training + validation), fold 2 (training + validation), fold 3 (training + validation), fold 4 (training + validation), and fold 5 (training + validation) are calculated. The mean MSE and standard deviation are calculated, and the performance metrics of these parameters are recorded. The mean MSE values ​​of all combined parameters are compared, the optimal kernel parameters are selected, and the final model is trained on the entire dataset using the optimal kernel parameters.

[0047] Furthermore, in the step of establishing a multivariate nonlinear regression model to predict the mass transfer performance of the hypergravity-enhanced decarbonization process: S51: Dataset preprocessing; S52: Dataset classification; S53: Establish a multivariate nonlinear regression model; S54: Use the NLINFIT function to obtain the regression coefficients; S55: Predict the mass transfer performance of the supergravity-enhanced decarbonization process.

[0048] In this embodiment, the same dataset preprocessing and dataset separation methods as those used in the machine learning model establishment process described above are employed. The multivariate nonlinear regression model is a commonly used method for data prediction, capable of simulating the nonlinear relationship between independent and dependent variables, and can consider the influence of multiple independent variables. A multinomial regression model is selected, as follows: In the formula, a, b, c, d, e and f are regression coefficients.

[0049] The regression coefficients are obtained using the NLINFIT function in MATLAB, as follows: Prepare the data for the machine learning model described above. Write a MATLAB function to describe the nonlinear regression model, providing initial guesses. Pass the independent variable data, dependent variable data, model function handle, and initial parameter vector as input to the NLINFIT function. The function will return the estimated parameter values, which are the regression coefficients. After obtaining the regression coefficients, the fit must be assessed. Plot the residuals to observe whether they exhibit a random distribution and whether there are obvious patterns or trends. Plot a comparison between the predicted and actual values ​​to visually evaluate the model's fit.

[0050] Using the obtained regression coefficients, a multivariate nonlinear regression model was established to predict the mass transfer performance of the hypergravity-enhanced decarbonization process.

[0051] Furthermore, in the steps of comparing and evaluating machine learning models that predict the mass transfer performance of the hypergravity-enhanced decarbonization process: Two model parameters were selected as evaluation parameters, and the root mean square error of the coefficient of determination was used to comprehensively compare and evaluate the accuracy of the machine learning model and the multivariate nonlinear regression model in predicting the mass transfer performance during the hypergravity-enhanced decarbonization process. Coefficient of determination (R) 2 The calculation formula is as follows:

[0052] R 2 The larger the R value, the higher the prediction accuracy of the model; conversely, the smaller the R value, the lower the accuracy of the model's predictions. 2 The smaller the value, the lower the accuracy of the model's predictions; The formula for calculating the root mean square error (RMSE) is as follows: RMSE is a commonly used indicator to measure the difference between predicted and actual values. The smaller the value, the better the model's prediction performance; conversely, the larger the value, the worse the model's prediction performance.

[0053] In this embodiment, the R-squared values ​​of the machine learning model and the multivariate nonlinear regression model are calculated. 2And RMSE values, combined with R 2 Using RMSE calculation data, machine learning models and multivariate nonlinear regression models are evaluated and compared.

[0054] Furthermore, in the step of predicting the mass transfer performance of the hypergravity-enhanced decarbonization process: By comprehensively comparing the evaluation coefficient R of machine learning models and multiple nonlinear regression models, 2 Using RMSE data, the accuracy of the machine model was verified, and the optimized machine model was used to predict the mass transfer performance during the hypergravity-enhanced decarbonization process.

[0055] For implementation example one, please refer to [link / reference]. Figures 7-9 .

[0056] Preprocessing of mass transfer performance experimental data during hypergravity-enhanced decarburization: Dimensionless analysis was used to preprocess the parameters affecting mass transfer performance during the hypergravity-enhanced decarburization process. The parameters were: gas Reynolds number (ReG), which is the ratio of gas viscous force to inertial force; liquid Reynolds number (ReL), which is the ratio of liquid viscous force to inertial force; liquid Schmidt number (ScL), which is the ratio of liquid viscosity coefficient to diffusion coefficient; gas Grashov number (GrG), which is the ratio of centrifugal force to viscous force; and liquid-gas molar ratio (Mr), which is the ratio of liquid molar mass to gas molar mass concentration. The Sherwood number (Sh) is the ratio of mass transfer coefficient to gas diffusion coefficient and packing specific surface area. The processed model input and output variables and their value ranges are shown in Table 1.

[0057] Table 1. Model input and output variables and their value ranges.

[0058] Screening of key parameters affecting mass transfer performance during hypergravity-enhanced decarburization: Since there are five input variables affecting mass transfer performance, involving numerous influencing parameters, to reduce the number of input variables and improve the prediction accuracy of the machine learning model, it is necessary to filter out and eliminate input variables with minimal impact on the output variables. The grey relational coefficient, Pearson correlation coefficient, Kendall correlation coefficient, and Spearman correlation coefficient are used to calculate the input variables. , , , , With output variables The degree of correlation, and the results of correlation coefficient calculations using different evaluation methods are shown in Table 2 and... Figure 7 :

[0059] Table 2. Evaluation Table of the Correlation Between Input and Output Variables According to Table 2 and Figure 7 It can be seen that Kendall's correlation coefficient is completely consistent with Spearman's and Pearson's in terms of positive and negative correlation, and can better reflect the relationship between each influencing parameter and the output variable. , , , It is positively correlated with the output variable. The input variables are negatively correlated with the output variables. The grey relational coefficients between all five input and output variables are greater than 0.5. Therefore, to improve prediction accuracy, , , , and All of them are used as input variables for model training.

[0060] Establishment of a machine learning model for predicting mass transfer performance in hypergravity-enhanced decarbonization processes: Use the randperm function to , , , and The dataset is randomly distributed in a 7:3 ratio, and the input features and output target variables are mapped to the [0,1] interval using the max-min normalization method; The radial basis function is chosen as the kernel parameter, and the regularization parameter γ takes values ​​in the range [0.1, 100], with an initial value of γ = 10; kernel parameters The value range is [0.01, 100], and the initial value is... =1. The optimal hyperparameter combination was automatically found using a grid search combined with 5-fold cross-validation. The results are shown in Table 3.

[0061] Table 3 Hyperparameter Performance Evaluation Table As shown in Table 3, the optimal hyperparameter combination is found using a grid search combined with 5-fold cross-validation. ]=[ ].

[0062] All datasets were trained using optimal hyperparameters, and the trained machine learning model was used to predict the mass transfer performance during hypergravity-enhanced decarbonization. Results are shown in [Table missing]. Figure 8 .

[0063] Establishment of a multivariate nonlinear regression model for predicting mass transfer performance in hypergravity-enhanced decarbonization processes: Based on experimental data, a multivariate nonlinear relationship model between five input variables and one output variable is established as follows: In the formula, there are a total of 6 regression coefficients to be determined; The regression coefficients were obtained using the NLINFIT function in MATLAB software, in the following order: , , , , , .

[0064] Using the obtained regression coefficients, a multivariate nonlinear regression prediction model for mass transfer performance during hypergravity-enhanced decarbonization was established. The prediction results are shown in [reference needed]. Figure 9 .

[0065] Comparison of prediction models for mass transfer performance in hypergravity-enhanced decarbonization processes: For machine learning models, the prediction results on the training set are: = 0.0055, = 0.9808; the prediction result for the test set is = 0.0096, = 0.9731; For the multivariate nonlinear regression model, the prediction result for the training set is: = 0.0116, = 0.9046; the prediction result for the test set is =0.0129, = 0.8954. This indicates that the machine learning model outperforms the multivariate nonlinear regression model in prediction. Therefore, a machine learning model is used to predict the mass transfer performance in the hypergravity-enhanced decarbonization process.

[0066] Predict the mass transfer performance of the decarbonization process enhanced by supergravity.

[0067] Using the optimized machine learning model, input , , , and Given a dataset of input variables, predict the corresponding Sh value, then perform inverse normalization on the predicted values ​​to achieve machine learning model prediction of mass transfer performance during hypergravity-enhanced decarbonization.

[0068] The above-disclosed embodiments are merely one or more preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes for implementing the above embodiments and equivalent changes made in accordance with the claims of this application still fall within the scope of this application.

Claims

1. A machine learning-based method for predicting the performance of enhanced carbon transfer in hypergravity, comprising the following steps: Preprocessing of experimental data on mass transfer performance during hypergravity-enhanced decarbonization; Screening of key parameters affecting mass transfer performance during hypergravity-enhanced decarbonization; Establish a machine learning model to predict the mass transfer performance of the hypergravity-enhanced decarbonization process; A multivariate nonlinear regression model was established to predict the mass transfer performance of the hypergravity-enhanced decarbonization process. Comparative evaluation of predictive models for mass transfer performance in hypergravity-enhanced decarbonization processes; Predict the mass transfer performance of the decarbonization process enhanced by supergravity.

2. The machine learning-based method for predicting decarbonization and mass transfer performance under hypergravity as described in claim 1, characterized in that, In the preprocessing steps for the mass transfer performance experimental data during the hypergravity-enhanced decarburization process, the preprocessing procedure is as follows: The variable parameters affecting the mass transfer performance of the supergravity-enhanced decarbonization process were identified, mainly including equipment structural parameters, gas-liquid physical property parameters, and operating parameters; The main structural parameters include: inner radius, outer radius, average radius, hydraulic diameter, thickness, porosity, and specific surface area of ​​the packing material; The main gas-liquid physical properties include: dynamic viscosity of the gas-liquid phase and diffusion coefficient of CO2 in the gas-liquid phase; The main operating parameters include: gas-liquid phase flow rate, gas-liquid phase apparent velocity, and packing rotation speed; The above-mentioned influencing parameters were transformed into dimensionless variables using a dimensionless analysis method.

3. The machine learning-based method for predicting decarbonization and mass transfer performance under hypergravity as described in claim 2, characterized in that, In the step of transforming the above influencing parameters into dimensionless variables using dimensionless analysis, the dimensionless analysis process is as follows: The gas Reynolds number is the ratio of viscous force to inertial force of the gas; the liquid Reynolds number is the ratio of viscous force to inertial force of the liquid; the liquid Schmidt number is the ratio of viscosity coefficient to diffusion coefficient of the liquid; the gas Grashov number is the ratio of centrifugal force to viscous force; and the liquid-gas molar ratio is the ratio of the molar mass of the liquid to the molar mass concentration of the gas. The Sherwood number is the ratio of the mass transfer coefficient to the gas diffusion coefficient and the specific surface area of ​​the packing.

4. The machine learning-based method for predicting decarbonization and mass transfer performance under hypergravity as described in claim 1, characterized in that, In the step of screening the main parameters affecting mass transfer performance during the ultragravity-enhanced decarburization process, the screening process is as follows: Four multi-factor correlation statistical analysis methods were used to assess the degree of correlation between the main influencing parameters and mass transfer performance; Grey relational coefficient analysis, Pearson coefficient analysis, Kendall coefficient analysis, and Spearman coefficient analysis were used to evaluate the correlation between influencing parameters and mass transfer performance. Input variables are filtered based on their correlation coefficients, eliminating those with low correlation, to determine the latest number of input variables for the machine learning model.

5. The machine learning-based method for predicting decarbonization and mass transfer performance under hypergravity as described in claim 1, characterized in that, In the steps of establishing a machine learning model to predict the mass transfer performance of the hypergravity-enhanced decarbonization process, the establishment process is as follows: Data preparation and partitioning: The randperm function is used to randomly partition the collected raw dataset into a training set and a test set. The training set is used for model training and parameter optimization, while the test set is used for model performance verification. Specifically, the total samples are randomly allocated according to a certain proportion to ensure the representativeness of the data distribution. Data normalization is performed to eliminate the impact of differences in the scale of different features on the model. The maximum-minimum normalization method is used to map the input features and the output target variable to the [0,1] interval. This process includes: normalizing the training set data and recording the normalization parameters; the test set data is transformed using the same normalization parameters as the training set to maintain the consistency of data processing. Model initialization, selecting radial basis functions to handle mapping relationships, and providing the value ranges of regularization parameters and kernel function parameters; Hyperparameter optimization employs a grid search combined with k-fold cross-validation to automatically find the optimal hyperparameter combination and determine the optimal parameter values ​​by minimizing the mean square error of cross-validation. Model training involves using optimized hyperparameters to train the model based on the training set data, and solving for the model parameters α Lagrange multiplier vector (α) and bias term (b). Model prediction and inverse normalization: The trained model is used to predict the test set to obtain normalized prediction values; then, the inverse transformation of the normalized parameters of the training set is used to restore the prediction results to the original dimensional space to obtain the final prediction values.

6. The machine learning-based method for predicting decarbonization and mass transfer performance under hypergravity as described in claim 5, characterized in that, In the step of optimizing hyperparameters using a grid search combined with k-fold cross-validation, the optimization process is as follows: Calculate the mean and standard deviation for each feature; Evolutionary standardization is applied to each eigenvalue; A logarithmic scale is used to ensure full coverage of the parameter space; The system iterates through all parameter combinations; The dataset is split into 5-fold partitions; Cross-validation was performed on each set of data, and the mean squared error and standard deviation were calculated. Record the performance metrics for each set of kernel parameters; Choose the optimal kernel parameters and train the model on the complete dataset using the optimal kernel parameters; Predict the mass transfer performance of the decarbonization process enhanced by supergravity.

7. The machine learning-based method for predicting decarbonization and mass transfer performance under hypergravity as described in claim 1, characterized in that, In establishing a multivariate nonlinear regression model to predict the mass transfer performance of the hypergravity-enhanced decarbonization process, the process is as follows: Data preprocessing; Dataset classification; Establish a multivariate nonlinear regression model; The regression coefficients are obtained using the NLINFIT function; Predict the mass transfer performance of the decarbonization process enhanced by supergravity.

8. The machine learning-based method for predicting decarbonization and mass transfer performance under hypergravity as described in claim 1, characterized in that, In the comparative evaluation of the predictive model steps for the mass transfer performance of the hypergravity-enhanced decarbonization process, the comparative evaluation process is as follows: Two model parameters are selected for evaluation: coefficient of determination and root mean square error, to evaluate the accuracy of the machine learning model. The closer the coefficient of determination is to 1, the higher the model accuracy; the closer it is to 0, the lower the model accuracy. This is suitable for evaluating the overall predictive accuracy of the model. The closer the mean relative error, root mean square error, and standard deviation are to 0, the better the model's predictive performance. The machine learning model and the multivariate nonlinear regression model are compared and evaluated using training and test set data to further assess the accuracy of the machine learning model. If the prediction results of the machine learning model are better than those of the multivariate nonlinear regression model, the optimized machine learning model can be used to predict the mass transfer performance in the hypergravity-enhanced decarbonization process.

Images