Carbon footprint prediction method for steel enterprises based on ensemble algorithm

Abstract

The steel enterprise carbon footprint prediction method based on integrated algorithm belongs to the technical field of carbon emission prediction.The present application uses K nearest neighbor algorithm, combines horizontal processing method and vertical processing method for data preprocessing, adopts Pearson correlation analysis and Copula theory to study the correlation between industrial system loads, determines the input characteristic variables, then uses Shapley value method to assign optimal weight coefficient to Stacking integrated model, and carries out carbon emission prediction through improved Stacking integrated algorithm.The prediction method analyzes the relationship between steel industry characteristic variables by using Person correlation coefficient and Copula theory, comprehensively analyzes the influence factors closely related to steel industry carbon footprint, and effectively improves the prediction accuracy of industrial carbon footprint of steel enterprises according to the improved Stacking integrated learning model.

Description

Technical Field

[0001] This invention belongs to the field of carbon emission prediction technology, and in particular relates to a method for predicting the carbon footprint of steel enterprises based on an integrated algorithm. Background Technology

[0002] In recent years, global climate change has become increasingly serious, primarily due to rising global temperatures caused by emissions of greenhouse gases such as carbon dioxide. Mitigating climate change has become a common global goal.

[0003] Research on carbon emission prediction methods has attracted increasing attention, as it plays a crucial guiding role in optimizing and formulating emission reduction policies. Existing carbon emission prediction research mainly falls into three categories: statistical models, artificial intelligence models, and hybrid models. However, most current research methods suffer from serious problems, such as difficulty in accurately modeling the relationship between emissions and influencing factors, insufficient real-time monitoring capabilities for greenhouse gas emissions, and low accuracy.

[0004] Real-time industrial carbon footprint (ICF) calculation plays a crucial role in addressing the aforementioned issues. Accurate ICF calculations help people understand a factory's emissions status, providing a basis for emission reduction. Simultaneously, real-time ICF calculations can improve the credibility of carbon emission disclosure data, enhance corporate social responsibility, and encourage companies to optimize their carbon reduction strategies. This is particularly important for steel companies with large carbon emissions. Currently, the accuracy of carbon footprint calculations for steel companies is low, and the prediction accuracy is poor.

[0005] Therefore, there is an urgent need for a new technical solution to address this problem. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a carbon footprint prediction method for steel enterprises based on an integrated algorithm to solve the current technical problems of low accuracy in carbon footprint calculation and poor prediction accuracy for steel enterprises.

[0007] The carbon footprint prediction method for steel enterprises based on ensemble algorithms includes the following steps, which are performed sequentially:

[0008] S1: Based on the carbon emission production process of steel enterprises, determine the carbon emission sources, which include electricity, various fuel energy sources and industrial raw materials related to carbon emissions;

[0009] The original dataset is obtained by acquiring the historical electricity consumption data of the enterprise to be predicted, the consumption data of various fuel energy in the production process, the consumption data of industrial raw materials related to carbon emissions, and the corresponding historical carbon emissions.

[0010] S2: The KNN algorithm is used to detect outliers in the electricity consumption, fuel energy consumption, industrial raw material consumption, and carbon emissions data in the original dataset, fill in the missing carbon emission data, and remove outliers; the corresponding electricity consumption, fuel energy consumption, industrial raw material consumption, and carbon emissions data are standardized by date to ensure that data of the same type are within the same magnitude range and form the original sample set.

[0011] S3: Using Person correlation analysis on the standardized original sample set, the correlation between electricity consumption, various fuel energy sources or industrial raw materials and carbon emissions is obtained. Energy sources or raw materials with high correlation are selected as feature variables according to a set threshold. Then, the optimal Copula function is used to calculate the correlation metric between each feature quantity to analyze the degree of correlation. Finally, based on the analysis results and the set threshold, the input feature variables are determined, namely the types of fuel energy, industrial raw materials, and electricity used by the steel company in carbon footprint prediction. The corresponding dataset is selected from the original sample set as the training sample set, where electricity consumption, fuel energy consumption data, and industrial raw material consumption data related to carbon emissions are used as feature values, and carbon emissions are used as the target value. The training sample set is divided into a training set and a test set.

[0012] S4: A two-layer Stacking ensemble learning model is adopted. The number and types of the first-layer base learners and the types of the second-layer meta learners of the Stacking ensemble learning model are selected. The first-layer base learners are trained based on the feature values ​​and target values ​​of the sample data in the training set to obtain a new dataset.

[0013] S5: The meta-learner of the Stacking ensemble learning model takes the new dataset obtained from the base learner and the training set as input, and uses the mean absolute percentage error as the index function to iteratively train and obtain the trained Stacking model.

[0014] S6: Use the feature values ​​of the sample data in the training set to predict carbon emissions through the trained Stacking model, obtain the predicted carbon emission values, calculate the Shapley value of each base model, assign weight coefficients according to these values, and obtain the weighted carbon emission prediction results; integrate the weighted carbon emission prediction results through the meta-model to form the final carbon emission prediction results, and obtain an improved Stacking ensemble learning model based on the trained Stacking model.

[0015] S7: Input the real-time electricity consumption and various fuel energy consumption during the production process into the improved Stacking ensemble learning model to obtain real-time carbon emission prediction results and complete the carbon footprint prediction of steel enterprises.

[0016] In step S2, the missing carbon emission data is filled using horizontal and vertical processing methods, with the specific formulas as follows:

[0017]

[0018] Where p(d-1,h) represents the carbon emission value in the same month as the anomaly in the previous year, p(d-1,h-1) is the carbon emission value in the month before the anomaly in the previous year, and p(d,h-1) is the carbon emission value in the month before the anomaly.

[0019] The formula for standardizing the corresponding electricity consumption, fuel energy consumption data, industrial raw material consumption data, and carbon emissions in step S2 is as follows:

[0020]

[0021] In the formula, x j Let j represent each influencing factor of the data sample, where j = 1, 2, ..., n; This represents the i-th value under the original influence factor; This represents the normalized value.

[0022] The formula for the Person correlation analysis in step S3 is:

[0023]

[0024] In the formula, r jk Let x be the Pearson correlation coefficient between two data variables. ij and x ik These represent the i-th data point in the j-th and k-th class of sample data, respectively. and and represent the mean values ​​of the sample data in class j and class k, respectively.

[0025] In step S3, the optimal Copula function is the distribution function and density function of the bivariate Frank Copula function. The optimal function is selected by calculating the average Euclidean distance of different Copula functions. Then, the optimal Copula function is used to calculate the correlation metric between each feature to determine the degree of association. The expressions are as follows:

[0026]

[0027]

[0028] Where k represents the correlation parameter, which controls the degree of dependence between two random variables. A larger k value indicates a stronger correlation, while a smaller k value indicates a weaker correlation. It is the distribution function of the bivariate Frank Copula function. It is the density function of the binary Frank Copula function, where u and v are the cumulative distribution function values ​​of the two random variables.

[0029] In step S4, the first layer of the Stacking ensemble learning model uses five types of base learners: lightweight gradient booster, random forest, K-nearest neighbor, extreme gradient booster, and support vector machine.

[0030] In step S4, the meta-learner uses the ridge regression algorithm.

[0031] The carbon footprint prediction system for steel enterprises based on ensemble algorithms adopts the aforementioned carbon footprint prediction method for steel enterprises based on ensemble algorithms, including a data processing module, a Stacking model construction module, and a Shapley value prediction module.

[0032] The data processing module is used to detect outliers in the original dataset, fill in missing data, and remove outliers; it is used to standardize the data in the original sample set; and it is also used to analyze the degree of correlation by calculating the correlation measure between each feature quantity through Pearson correlation coefficient and optimal Copula function, and to determine the input feature variables.

[0033] The Stacking model building module is used to build a Stacking ensemble learning model and initialize the number and types of base learners and the types of meta learners in the Stacking ensemble learning model.

[0034] The Shapley value module is used to calculate the Shapley value of each base model and assign weight coefficients based on these values ​​to obtain a weighted carbon emission prediction result; it is used to integrate the weighted carbon emission prediction results through a meta-model to form the final carbon emission prediction result.

[0035] Through the above design scheme, the present invention can bring the following beneficial effects:

[0036] 1. This invention addresses the load data characteristics of steel enterprises by employing a multi-scheme data preprocessing method. First, it uses the K-nearest neighbor algorithm, horizontal processing, and vertical processing to perform data detection. Then, based on the characteristics of the steel enterprise datasets, it uses Copula theory combined with the characteristics of different datasets to conduct correlation analysis. This innovative method provides theoretical support for calculating the carbon footprint of steel enterprises.

[0037] 2. This invention innovatively proposes an improved Stacking ensemble model, specifically applied to energy data from steel enterprises. By utilizing the Shapley value method in cooperative game theory to assign optimal weight coefficients to multiple prediction models, and with shorter model training and update times, an optimal ensemble prediction model for the carbon footprint of steel enterprises is achieved, effectively improving prediction accuracy and practicality.

[0038] 3. This invention innovatively proposes a carbon footprint calculation method for steel enterprises based on energy conversion methods. This method leverages the strong correlation between electricity consumption and fuel energy in steel production, using current electricity consumption and related energy data to drive the prediction of current fuel energy consumption, and then converts the predicted fuel energy consumption and electricity consumption into corresponding carbon emissions. Compared with traditional industrial carbon footprint calculation methods, this invention demonstrates more accurate and efficient prediction performance. Attached Figure Description

[0039] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0040] Figure 1 The flowchart shows the carbon footprint prediction method for steel enterprises based on integrated algorithms according to the present invention.

[0041] Figure 2 This is a block diagram illustrating the principle of the Stacking ensemble learning model in an example of the present invention;

[0042] Figure 3 This is a carbon footprint prediction diagram of the conventional prediction method and the energy conversion method in the examples of this invention;

[0043] Figure 4 This is a carbon footprint prediction diagram of the conventional prediction method and the energy conversion method in the examples of this invention. Detailed Implementation

[0044] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0045] A flowchart of a prediction process based on an ensemble model using the Shapley value method is shown below. Figure 1 As shown, its carbon emission prediction method includes the following steps:

[0046] S1: Datasets were obtained from industrial scenarios of multiple steel companies, covering historical power load, fuel energy, carbon emission-related industrial raw materials, and carbon emissions of the industrial system from January 2011 to December 2020. Fuel energy includes one or more of coke, raw coal, coal tar, diesel, liquefied petroleum gas, and natural gas; carbon emission-related industrial raw materials include crude steel, etc. The entire industrial dataset was divided into training and testing sets in an 8:2 ratio, and predictions were made for each classified fuel energy, carbon emission-related industrial raw material, and carbon emissions.

[0047] S2: Preprocess the sample data of the dataset, specifically including the following steps:

[0048] The K-nearest neighbor algorithm is used to identify outliers in historical data. The K-nearest neighbor algorithm is used to identify outliers and missing values ​​in the original dataset, and then these data are removed. Data imputation is performed by combining horizontal and vertical processing methods to ensure the integrity of the dataset. The min-max normalization method is used to normalize the original dataset so that the values ​​of all influencing factors in the original dataset, including historical carbon emission data, are in the range [0,1].

[0049] The missing carbon emission data was filled using horizontal and vertical processing methods, with the specific formulas as follows:

[0050]

[0051] Where p(d-1,h) represents the carbon emission value in the same month as the anomaly in the previous year, p(d-1,h-1) is the carbon emission value in the month before the anomaly in the previous year, and p(d,h-1) is the carbon emission value in the month before the anomaly.

[0052] The formula for standardizing the corresponding electricity consumption, fuel energy consumption data, carbon emission-related industrial raw material consumption data, and carbon emissions using the min-max standardization method is as follows:

[0053]

[0054] In the formula, x j Let j represent each influencing factor of the data sample, where j = 1, 2, ..., n; This represents the i-th value under the original influence factor; This represents the normalized value.

[0055] S3: Use Pearson correlation analysis and Copula theory to obtain the correlation between features and determine the input feature variables, which are the consumption of raw coal, coke, diesel, liquefied petroleum gas, natural gas, and tar, specifically:

[0056] S31: Use the Person correlation coefficient to obtain the correlation between features and preliminarily determine the input feature variables, which are coke, crude steel, electricity, raw coal, tar, diesel, liquefied petroleum gas, and natural gas.

[0057] S32: Marginal distribution estimation and joint distribution study of industrial energy load data; then, the optimal function is selected by calculating the average Euclidean distance of different Copula functions; finally, the correlation measure between various characteristic quantities is calculated based on the optimal Copula function to analyze the degree of correlation, and the expressions are as follows:

[0058]

[0059]

[0060] Here, k represents the correlation parameter, which controls the degree of dependence between the two random variables. A larger k value indicates a stronger correlation, while a smaller k value indicates a weaker correlation. It is the distribution function of the bivariate Frank Copula function. It is the density function of the binary Frank Copula function, where u and v are the cumulative distribution function values ​​of the two random variables.

[0061] S33: Based on the analysis results and the set thresholds, determine the input feature variables, which are raw coal, coke, diesel, liquefied petroleum gas, natural gas, and tar.

[0062] S4: Construct a two-layer stacking ensemble learning model, such as... Figure 2 As shown in Table 1, the Stacking ensemble learning model is configured with base learners and meta-learners. The first layer of the Stacking ensemble learning model is constructed by selecting five models from traditional statistical algorithm models and existing machine learning algorithm models.

[0063] Table 1. Stacking Algorithm Prediction Model

[0064]

[0065] The initialization of the first layer of the Stacking ensemble learning model involves initializing the parameters of the base learner and meta-learner networks as follows:

[0066] S41: Set parameters such as learning rate, tree depth, number of leaf nodes, and regularization strength for the lightweight gradient booster.

[0067] S42: Set parameters such as the number of decision trees in the random forest, the maximum depth of the trees, and the minimum number of samples required for a leaf node;

[0068] S43: Set parameters such as the number of nearest neighbors of K nearest neighbors and the weight allocation function for each neighbor;

[0069] S44: Set parameters such as the learning rate of the extreme gradient booster, the depth of the tree, the proportion of subsamples, and the regularization strength;

[0070] S45: Set parameters such as regularization coefficients, kernel function type, and kernel function coefficients for the support vector machine.

[0071] S5: The meta-learner of the Stacking ensemble learning model takes the new dataset obtained by the base learner and the training set as input, and uses the normalized dataset after combining them. Iterative training is performed with the mean absolute percentage error as the index function, so that the meta-learner learns the implicit relationship between the training set and the new dataset, and obtains the trained Stacking model.

[0072] To evaluate the predictive performance of the model, this invention employs two evaluation metrics: root mean square error (RMSE) and root mean square error (y). i ,y fi ) and mean absolute percentage error (MAPE(y)) i ,y fi The calculation expressions for the two error evaluation indicators are as follows:

[0073]

[0074]

[0075] In the formula, y i y is the actual value of the i-th sample point. fi is the predicted value of the i-th sample point, and m is the number of sample points;

[0076] Ultimately, in order to improve the accuracy of model predictions, this invention uses the average absolute percentage error of the model output as the optimization target. After iterative training is completed, a trained model is obtained. The test set is then substituted to obtain the carbon emission prediction value.

[0077] S6: Establish multiple algorithm models for integrated prediction. Calculate the Shapley value for each base model and assign weight coefficients based on these values. The weights reflect the contribution of each base model to the overall prediction. Integrate the prediction results using the weights of the base models to form a weighted meta-feature vector. Each element in this vector is a base model prediction adjusted based on the Shapley value, effectively fusing information from different models. Finally, the meta-model uses the weighted meta-feature vector for prediction, generating the final carbon emission prediction result.

[0078] To verify the advantages of the proposed ensemble model based on the Shapley value method, three prediction schemes are proposed for comparison based on the idea of ​​controlled variables, thereby demonstrating the effectiveness and feasibility of the prediction scheme of this invention. Taking the historical power load, fuel energy, and carbon emission data of the industrial systems of multiple steel companies from January 2011 to December 2020 as examples, the following comparison of three industrial carbon footprint prediction schemes for steel enterprises is presented:

[0079] (1) Prediction scheme 1: Use a single prediction model to make predictions and obtain the prediction results of total industrial carbon emissions;

[0080] (2) Prediction scheme 2: The total carbon emissions are predicted by using the conventional Stacking ensemble model.

[0081] (3) Prediction scheme 3: The total industrial carbon emissions are predicted using an improved Stacking ensemble model based on the Shapley value method, and the prediction results of the total industrial carbon emissions of steel enterprises are obtained by superimposing the data.

[0082] To clearly compare the three load forecasting schemes, the methods used for each scheme are illustrated in Table 2.

[0083] Table 2 Comparison of Three Carbon Emission Prediction Schemes

[0084]

[0085] To more intuitively compare the results of the three prediction schemes, carbon emission prediction curves for the three schemes are plotted. The prediction results for the three different types of industrial carbon footprints are shown below. Figure 3 and Figure 4 As shown.

[0086] The root mean square error and mean absolute percentage error of the three prediction schemes can be calculated from their outputs. The error index comparison is shown in Table 3.

[0087] Table 3 Comparison of errors among the three prediction schemes

[0088]

[0089] Compared to prediction scheme 1, prediction scheme 2 has a higher RMSE (18.50) than prediction scheme 1 (M2), which has the lowest RMSE among schemes 1, but a lower RMSE (18.52) than the average RMSE of scheme 1 (227.12). In terms of MAPE (MAPE), it is 0.576% lower than prediction scheme 1 (M3), which has the lowest MAPE among schemes 1. This indicates that, without considering optimization based on the Shapley method, the ensemble model in this paper outperforms most single prediction models, but its prediction performance is somewhat lacking compared to a few high-performing single models.

[0090] Among them, prediction scheme 3, proposed in this invention, reduces RMSE by 25.80, MAPE by 1.079%, and improves prediction accuracy by 12.49% compared with prediction scheme 2. It can be seen that compared with the ordinary ensemble model, the proposed improved ensemble model method based on Shapley value method can effectively improve the load prediction effect.

[0091] Compared to prediction scheme 1, prediction scheme 3 shows the following improvements: In terms of RMSE, it is 6.301 lower than the lowest RMSE prediction scheme 1 (M2) and 56.90 lower than the highest RMSE prediction scheme 1 (M1); in terms of MAPE, it is 1.155% lower than the lowest MAPE prediction scheme 1 (M3), with a prediction accuracy improvement of 22.17%, and 3.029% lower than the highest MAPE prediction scheme 1 (M5), with a prediction accuracy improvement of 42.76%. It can be seen that the prediction accuracy improvement of this invention is the most significant, validating the effectiveness of combining the improved Shapley value method with the Stacking ensemble model, further enhancing load forecasting performance.

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, utilizing the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A carbon footprint prediction method for steel enterprises based on ensemble algorithms, characterized by: The steps include the following steps, and the following steps are performed in sequence: S1: Based on the carbon emission production process of steel enterprises, determine the carbon emission sources, which include electricity, various fuel energy sources and industrial raw materials related to carbon emissions; The original dataset is obtained by acquiring the historical electricity consumption data of the enterprise to be predicted, the consumption data of various fuel energy in the production process, the consumption data of industrial raw materials related to carbon emissions, and the corresponding historical carbon emissions. S2: The KNN algorithm is used to detect outliers in the electricity consumption, fuel energy consumption, industrial raw material consumption, and carbon emissions data in the original dataset, fill in the missing carbon emission data, and remove outliers; the corresponding electricity consumption, fuel energy consumption, industrial raw material consumption, and carbon emissions data are standardized by date to ensure that data of the same type are within the same magnitude range and form the original sample set. S3: Using Person correlation analysis on the standardized original sample set, the correlation between electricity consumption, various fuel energy sources or industrial raw materials and carbon emissions is obtained. Energy sources or raw materials with high correlation are selected as feature variables according to a set threshold. Then, the optimal Copula function is used to calculate the correlation metric between each feature quantity to analyze the degree of correlation. Finally, based on the analysis results and the set threshold, the input feature variables are determined, namely the types of fuel energy, industrial raw materials, and electricity used by the steel company in carbon footprint prediction. The corresponding dataset is selected from the original sample set as the training sample set, where electricity consumption, fuel energy consumption data, and industrial raw material consumption data related to carbon emissions are used as feature values, and carbon emissions are used as the target value. The training sample set is divided into a training set and a test set. S4: A two-layer Stacking ensemble learning model is adopted. The number and types of the first-layer base learners and the types of the second-layer meta learners of the Stacking ensemble learning model are selected. The first-layer base learners are trained based on the feature values ​​and target values ​​of the sample data in the training set to obtain a new dataset. S5: The meta-learner of the Stacking ensemble learning model takes the new dataset obtained by the base learner and the training set as input, and uses the mean absolute percentage error as the index function to iteratively train and obtain the trained Stacking model. S6: Use the feature values ​​of the sample data in the training set to predict carbon emissions through the trained Stacking model, obtain the predicted carbon emission values, calculate the Shapley value of each base model, assign weight coefficients according to these values, and obtain the weighted carbon emission prediction results; integrate the weighted carbon emission prediction results through the meta-model to form the final carbon emission prediction results, and obtain an improved Stacking ensemble learning model based on the trained Stacking model. S7: Input the real-time electricity consumption and various fuel energy consumption during the production process into the improved Stacking ensemble learning model to obtain real-time carbon emission prediction results and complete the carbon footprint prediction of steel enterprises.

2. The carbon footprint prediction method for steel enterprises based on ensemble algorithms according to claim 1, characterized in that: In step S2, the missing carbon emission data is filled using horizontal and vertical processing methods, with the specific formulas as follows: Where p(d-1,h) represents the carbon emission value in the same month as the anomaly in the previous year, p(d-1,h-1) is the carbon emission value in the month before the anomaly in the previous year, and p(d,h-1) is the carbon emission value in the month before the anomaly.

3. The carbon footprint prediction method for steel enterprises based on ensemble algorithms according to claim 1, characterized in that: The formula for standardizing the corresponding electricity consumption, fuel energy consumption data, industrial raw material consumption data, and carbon emissions in step S2 is as follows: In the formula, x j Let j represent each influence factor of the data sample, where j = 1, 2, ..., n; This represents the i-th value under the original influence factor; This represents the normalized value.

4. The carbon footprint prediction method for steel enterprises based on ensemble algorithms according to claim 1, characterized in that: The formula for the Person correlation analysis in step S3 is: In the formula, r jk Let x be the Pearson correlation coefficient between two data variables. ij and x ik These represent the i-th data point in the j-th and k-th class of sample data, respectively. and represent the mean values ​​of the sample data in class j and class k, respectively.

5. The carbon footprint prediction method for steel enterprises based on ensemble algorithms according to claim 1, characterized in that: In step S3, the optimal Copula function is the distribution function and density function of the bivariate Frank Copula function. The optimal function is selected by calculating the average Euclidean distance of different Copula functions. Then, the optimal Copula function is used to calculate the correlation metric between each feature to determine the degree of association. The expressions are as follows: Where k represents the correlation parameter, which controls the degree of dependence between two random variables. A larger k value indicates a stronger correlation, while a smaller k value indicates a weaker correlation. It is the distribution function of the bivariate Frank Copula function; It is the density function of the binary Frank Copula function; u and v are the cumulative distribution function values ​​of the two random variables, respectively.

6. The carbon footprint prediction method for steel enterprises based on ensemble algorithms according to claim 1, characterized in that: In step S4, the first layer of the Stacking ensemble learning model uses five types of base learners: lightweight gradient booster, random forest, K-nearest neighbor, extreme gradient booster, and support vector machine.

7. The carbon footprint prediction method for steel enterprises based on ensemble algorithms according to claim 1, characterized in that: In step S4, the meta-learner uses the ridge regression algorithm.

8. A carbon footprint prediction system for steel enterprises based on an ensemble algorithm, employing the carbon footprint prediction method for steel enterprises based on an ensemble algorithm as described in claim 1, characterized in that: It includes a data processing module, a Stacking model building module, and a Shapley value prediction module; The data processing module is used to detect outliers in the original dataset, fill in missing data, and remove outliers; it is used to standardize the data in the original sample set; and it is also used to analyze the degree of correlation by calculating the correlation measure between each feature quantity through Pearson correlation coefficient and optimal Copula function, and to determine the input feature variables. The Stacking model building module is used to build a Stacking ensemble learning model and initialize the number and types of base learners and the types of meta learners in the Stacking ensemble learning model. The Shapley value method module is used to calculate the Shapley value of each base model and assign weight coefficients based on these values ​​to obtain weighted carbon emission prediction results. This is used to integrate the weighted carbon emission prediction results through a meta-model to form the final carbon emission prediction result.

Images