A method and system for predicting and verifying combustion carbon emission factors of gas-fired generator sets

Abstract

The present invention relates to the field of low-carbon, energy-saving and environmentally friendly technology, and discloses a method and system for predicting and verifying the carbon emission factor of combustion of a gas-fired power generation unit. By performing sensitivity and feature importance analysis on the gas component parameters of a gas sample, a comprehensive importance evaluation index is established. According to the numerical values ​​of the comprehensive importance evaluation index of each gas component parameter, a random forest combined with backward sequence selection is adopted to establish a gas carbon emission factor prediction model. This effectively achieves the prediction and verification of the accuracy and rationality of the combustion carbon emission factor results of the gas-fired power generation unit, thereby effectively ensuring the quality of the carbon emission accounting data of the gas-fired power generation unit and significantly improving the reliability of the carbon emission accounting data. In addition, it also helps the carbon market to grasp the changing trend of the carbon emission factor of the gas-fired power generation unit, thereby providing a data basis for the carbon market quota allocation, benchmark value setting and price regulation.

Description

Technical Field

[0001] The present invention belongs to the technical field of low-carbon, energy-saving and environmental protection, and specifically relates to a method and system for predicting and verifying the combustion carbon emission factor of a gas-fired generator set. Background Art

[0002] Direct emissions from fuel combustion dominate the carbon emissions of gas-fired power plants, typically accounting for over 90% of total emissions. Since gas-fired power plants primarily rely on natural gas for power generation, their carbon emissions are highly dependent on fuel characteristics, combustion efficiency, and operating conditions. Therefore, the accuracy of the emission factor for the combustion phase of a gas-fired power plant directly determines the reliability of the carbon emissions accounting for the entire plant. Using inappropriate emission factor data can lead to systematic deviations in accounting results, impacting carbon market trading, emission reduction policymaking, and corporate carbon asset management.

[0003] As a key parameter linking fuel consumption and actual emissions, the accurate measurement of the carbon emission factor is crucial for carbon accounting of gas-fired generators. Therefore, establishing a scientific and dynamic calibration method for carbon emission factors for gas-fired generators is of great practical significance for improving carbon data quality, ensuring fairness in the carbon market, and optimizing emission reduction policies.

[0004] Currently, when calculating fossil fuel carbon emissions, most gas-fired power generation units generally use the default carbon content per unit calorific value provided in the accounting guidelines to calculate combustion carbon emission factors for operational convenience or to achieve lower carbon emissions. However, due to differences in natural gas composition from different gas sources and fluctuations in unit combustion efficiency, the use of a uniform default value may cause the accounting results to deviate from the actual emission level. This practice can also lead to a systematic underestimation of carbon emissions, making it difficult to ensure the accuracy of carbon emission accounting results.

[0005] Therefore, there is an urgent need for an accurate and efficient method and system for predicting and verifying the carbon emission factors of gas-fired power generation units to effectively ensure the quality of carbon emission accounting data and to objectively evaluate the reliability and rationality of carbon emission data reported by gas-fired power generation enterprises. Summary of the Invention

[0006] One of the purposes of the present invention is to provide a method for predicting and verifying the combustion carbon emission factor of a gas-fired power generation set. By performing sensitivity and characteristic importance analysis on the gas component parameters of the gas sample, a comprehensive importance evaluation index is established; according to the numerical size of the comprehensive importance evaluation index of each gas component parameter, a random forest combined with a backward sequence selection algorithm is used to establish a gas carbon emission factor prediction model; thereby effectively realizing the accuracy and rationality of the prediction and verification of the combustion carbon emission factor results of the gas-fired power generation set, effectively ensuring the quality of the carbon emission accounting data of the gas-fired power generation set, and significantly improving the reliability of the carbon emission accounting data.

[0007] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is as follows:

[0008] A method for predicting and verifying the carbon emission factor of a gas-fired generator set includes the following steps:

[0009] S1. Sample collection and processing: Collect a group of gas samples with known component data sets and obtain the gas component data sets of the group of gas samples; the gas component data sets include the gas component parameters, elemental carbon content and lower calorific value data of each gas sample; based on the elemental carbon content and lower calorific value data of each gas sample in the gas component data sets, calculate the gas carbon emission factor of each gas sample combustion; the gas samples are gas samples used for combustion in gas-fired generator sets, and the gas component data sets of the group of gas samples include methane (CH4), ethane (C2H6), propane (C3H8), isobutane (i-C4H 10 ), n-butane (n-C4H 10 ), isopentane (i-C5H 12 ), n-pentane (n-C5H 12 ), carbon dioxide (CO2), nitrogen (N2), as well as elemental carbon content (C), lower calorific value (NCV);

[0010] S2. Sensitivity Analysis: Analyze the sensitivity between each gas carbon emission factor in the group of gas samples and each gas component parameter in the gas component dataset to obtain the sensitivity coefficient between each gas component parameter and the gas carbon emission factor. Then, rank the sensitivity coefficients of each gas component parameter to analyze the degree of influence of each gas component parameter on the gas carbon emission factor. In other words, analyze the sensitivity between different gas component parameters of all coal-fired samples and the gas carbon emission factors of all coal-fired samples to determine the degree of influence of different gas component parameters on the gas carbon emission factor.

[0011] S3. Feature Importance Analysis: Calculate the feature importance of each gas component parameter and rank them by their importance. By comparing the feature importance of each gas component parameter, we can quantify its contribution to the gas carbon emission factor prediction model.

[0012] S4. Establish a comprehensive evaluation index of importance: The sensitivity coefficients and characteristic importance of each gas component parameter are integrated to establish a comprehensive evaluation index of the importance of gas component parameters, and comprehensively evaluate the importance of each gas component parameter to the gas carbon emission factor;

[0013] S5. Establish a prediction model for gas carbon emission factors: Based on the numerical values ​​of the comprehensive importance evaluation indicators of each gas component parameter, a random forest combined with a backward sequence selection algorithm is used to establish a prediction model for gas carbon emission factors;

[0014] S6. Data verification: The third-party verification agency inputs the various gas component parameters of the gas-fired power generation unit reported by the enterprise into the gas carbon emission factor prediction model established in step S5 to obtain the model prediction value, and then calculates the absolute value of the error between the model prediction value and the gas carbon emission factor data reported by the enterprise, and compares the absolute value of the error with the preset threshold: if the absolute value of the error is less than or equal to the preset threshold, the data is judged to be reasonable; if the absolute value of the error is greater than the preset threshold, the data is judged to be questionable and further verification of the various gas component parameter data of the gas-fired power generation unit is required.

[0015] Preferably, the sensitivity coefficient r between each gas component parameter and the gas carbon emission factor is calculated according to the following formula in S2:

[0016] ;

[0017] Where r i is the sensitivity coefficient of the i-th gas component parameter, I ij is the i-th gas component parameter of the j-th sample, O j is the gas carbon emission factor of the jth sample, n is the number of gas samples, and m is the number of gas component parameters of each gas sample;

[0018] The range of the sensitivity coefficient r is 0<r<1. The closer the value of the sensitivity coefficient of the gas component parameter is to 1, the stronger the correlation between the gas component parameter and the gas carbon emission factor is, that is, the change of the component parameter will significantly affect the result of the gas carbon emission factor; a value close to 0 means no linear correlation.

[0019] Preferably, in S3, the feature importance M of each gas component parameter is calculated using the random forest algorithm. For each gas component parameter, the original prediction error e is first calculated based on the trained random forest model using the out-of-bag samples corresponding to each decision tree. t , the original prediction error e t is the mean absolute error between the original gas carbon emission factor model prediction value and the actual value of the gas carbon emission factor; then the values ​​of the gas component parameters in the out-of-bag samples are randomly disrupted and the prediction error is recalculated ; Finally, the difference in prediction error before and after randomly disrupting the gas component parameters is calculated to obtain the characteristic importance M of the gas component parameters; the larger the characteristic importance M of a certain gas component parameter, the greater the impact of the gas component parameter on the performance of the gas carbon emission factor prediction model.

[0020] Preferably, in said S4, in the process of establishing the comprehensive evaluation index of the importance of the gas component parameters, the weight of the sensitivity coefficient of the gas component parameters is the same as the weight of the characteristic importance.

[0021] Preferably, in said S5, the method for establishing a fuel gas carbon emission factor prediction model includes:

[0022] S5.1. Data Preprocessing: Based on the set of gas samples with known gas composition parameters from step S1, a full-variable random forest regression model for the gas carbon emission factor is established using the random forest algorithm. The initial dataset of gas samples is preprocessed to obtain a preprocessed dataset. The data in the initial dataset consists of the raw measurement data of the gas samples. The raw measurement data includes the gas composition parameters, elemental carbon content, and lower heating value data of each gas sample. This raw measurement data serves as the raw data for the training samples to ensure the stability and generalization ability of the model. The gas composition data of the gas samples used for training must be standard gas data or gas data tested by a CMA-accredited or CNAS-certified testing agency / laboratory.

[0023] S5.2. Bootstrap Sampling: Bootstrap sampling is performed on the preprocessed dataset in the training set. That is, n samples are randomly drawn with replacement from the original preprocessed dataset N, forming a new training set. This process is repeated until the size of each new training set is the same as the size of the preprocessed dataset.

[0024] S5.3. Parameter Selection: Use grid search and cross-validation to determine the optimal parameters for random forest modeling. Then, based on the selected optimal parameters, determine the number of decision trees and the minimum number of leaves, and construct the decision tree.

[0025] S5.4. Decision Tree Construction: Build a decision tree using each new training set obtained through the self-sampling in step S5.2. During the decision tree construction process, for each node in the tree, instead of considering all features for optimal segmentation, a portion of the gas component parameter features is randomly selected from all gas component parameter features, and then the optimal segmentation feature is selected from these randomly selected features:

[0026] S5.5. Integrated prediction: After all decision trees are trained, a gas carbon emission factor prediction model is formed;

[0027] After the gas carbon emission factor prediction model is formed, the gas carbon emission factor prediction model uses each decision tree to predict the newly input sample data, and then averages all the prediction results to obtain the final model prediction value.

[0028] Preferably, the establishment of the gas carbon emission factor prediction model in S5 also includes: using a backward sequence selection algorithm combined with a full-variable random forest regression model, arranging the comprehensive importance evaluation indicators of the gas component parameters in descending order, and successively eliminating the single gas component parameters with the lowest current comprehensive importance evaluation index values, and reconstructing a new gas carbon emission factor prediction model.

[0029] Preferably, the S5 further includes using the mean absolute percentage error as a performance evaluation index MAPE of the gas carbon emission factor prediction model:

[0030] After each removal of a single gas component parameter with the lowest overall importance evaluation index, the rationality of feature removal is evaluated by calculating the change in the MAPE (Massive Performance Evaluation) performance indicator of the new gas carbon emission factor prediction model. The removal threshold is set at a relative increase in MAPE of no more than 10%. (That is, if the MAPE increase in the new model is ≤10%, performance is considered to have not significantly degraded, and the removed parameter is retained; otherwise, the removed parameter is restored.) This iterative backward screening gradually removes redundant gas component parameters until further removal of any gas component parameter results in a MAPE increase exceeding the 10% threshold. Finally, based on the prediction model performance evaluation index and the number of gas parameters, the gas component parameters most relevant to the gas carbon emission factor are comprehensively selected. The optimal gas carbon emission factor prediction model is constructed using these selected gas component parameters. Reducing irrelevant features can reduce the model's generalization error and improve its predictive capability.

[0031] Preferably, the step S5.1 comprises: normalizing the input and output data using a minimum-maximum normalization method to map the data to the interval [0, 1];

[0032] Step S5.3 includes: optimizing the core parameters of the number of decision trees (trees) and the minimum number of leaves (min_leaf); establishing a complete parameter search space by setting candidate ranges for the number of decision trees and the minimum number of leaves; calculating the predictive performance of each parameter combination using a 5-fold cross-validation strategy and mean-squared error (MSE) as the evaluation metric, and ultimately selecting the optimal parameter combination that minimizes the MSE. After parameter optimization is complete, the optimal parameters are fixed as the modeling configuration.

[0033] Preferably, the method further includes predictive model performance evaluation: using a set of test set samples to evaluate the predictive performance of the predictive model built in step S5.

[0034] Another object of the present invention is to provide a system for predicting and verifying the carbon emission factor of a gas-fired power generation unit, wherein the system executes the above method and comprises:

[0035] Data collection and processing module, importance comprehensive evaluation index generation module, prediction model establishment module, data verification module;

[0036] The data collection and processing module is used to collect and process samples, collect a group of gas samples with known component data sets, obtain a gas component data set of the group of gas samples, and the gas component data set includes gas component parameters, elemental carbon content, and lower calorific value data of each gas sample; and calculate the gas carbon emission factor of each gas sample according to the elemental carbon content and lower calorific value data of each gas sample in the gas component data set;

[0037] The importance comprehensive evaluation index generation module is used to perform sensitivity analysis and characteristic importance analysis on each gas component parameter in the group of gas samples, and integrate the sensitivity coefficient and characteristic importance of each gas component parameter to establish a comprehensive evaluation index of the importance of the gas component parameter, and comprehensively evaluate the importance of each gas component parameter to the gas carbon emission factor;

[0038] The prediction model establishment module is used to establish a gas carbon emission factor prediction model based on the numerical value of the comprehensive importance evaluation index of each gas component parameter using a random forest combined with a backward sequence selection algorithm;

[0039] The data verification module is used to input the various gas component parameters of the gas generator set reported by the enterprise into the gas carbon emission factor prediction model to obtain the model prediction value, calculate the absolute value of the error between the model prediction value and the gas carbon emission factor data reported by the enterprise, and determine whether the data reported by the enterprise is reasonable.

[0040] Beneficial effects:

[0041] The method and system for predicting and verifying the combustion carbon emission factor of a gas-fired power generation unit proposed in the present invention establish a comprehensive importance evaluation index by performing sensitivity and characteristic importance analysis on the gas component parameters of the gas sample; according to the numerical values ​​of the comprehensive importance evaluation indexes of the gas component parameters, a random forest combined with a backward sequence selection algorithm is adopted to establish a gas carbon emission factor prediction model; thereby effectively realizing the prediction and verification of the accuracy and rationality of the combustion carbon emission factor results of the gas-fired power generation unit, and thus effectively ensuring the quality of the carbon emission accounting data of the gas-fired power generation unit, and significantly improving the reliability of the carbon emission accounting data. In addition, it also helps the carbon market to grasp the changing trend of the carbon emission factor of the gas-fired power generation unit, thereby providing a data basis for the carbon market quota allocation, benchmark value setting and price regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Shown is a flow chart of a method for predicting and verifying the carbon emission factor of a gas-fired power generation unit;

[0043] Figure 2 The figure shows the results of the sensitivity analysis of gas component parameters in a specific implementation case of the method for predicting and verifying the carbon emission factor of gas-fired power generation units;

[0044] Figure 3 Shown is a graph showing the results of characteristic importance analysis of gas component parameters in a specific implementation case of the method for predicting and verifying the carbon emission factor of gas-fired power generation units. DETAILED DESCRIPTION

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings and other embodiments can be obtained based on these drawings without inventive work.

[0046] The technical solution of the present invention is described in detail below with reference to specific embodiments.

[0047] Example 1

[0048] refer to Figure 1 A method for predicting and verifying the carbon emission factor of a gas-fired generator set includes the following steps:

[0049] S1. Sample collection and processing: Collect a group of gas samples with known component data sets, obtain the gas component data sets of the group of gas samples, and the gas component data sets include the gas component parameters, elemental carbon content and lower calorific value data of each gas sample; according to the elemental carbon content and lower calorific value data of each gas sample in the gas component data sets, calculate the gas carbon emission factor of each gas sample combustion; the gas sample is the gas sample used for combustion of the gas generator set, and the gas component data sets of the gas sample include methane (CH4), ethane (C2H6), propane (C3H8), isobutane (i-C4H 10 ), n-butane (n-C4H 10 ), isopentane (i-C5H 12 ), n-pentane (n-C5H 12 ), carbon dioxide (CO2), nitrogen (N2), as well as elemental carbon content (C), lower calorific value (NCV);

[0050] Based on the elemental carbon content and calorific value of the above gas component dataset, the gas carbon emission factor is calculated using the following formula:

[0051] ;

[0052] Where, CEF is the carbon emission factor of the gas, in tCO2 / GJ; C is the elemental carbon content of the gas, in tC / t; NCV is the lower calorific value of the gas, in GJ / t; OF is the oxidation rate of the gas, in %. The carbon oxidation rate of the gas specified in the current accounting guidelines is 99%.

[0053] S2. Sensitivity Analysis: Analyze the sensitivity between each gas carbon emission factor in the gas sample set and each gas component parameter in the gas component dataset to obtain the sensitivity coefficients between each gas component parameter and the gas carbon emission factor. Rank the sensitivity coefficients of each gas component parameter to analyze the degree of influence of each gas component parameter on the gas carbon emission factor.

[0054] S3. Feature Importance Analysis: Calculate the feature importance of each gas component parameter and rank them by their importance. By comparing the feature importance of each gas component parameter, we can quantify its contribution to the gas carbon emission factor prediction model.

[0055] S4. Establish a comprehensive evaluation index of importance: The sensitivity coefficients and characteristic importance of each gas component parameter are integrated to establish a comprehensive evaluation index of the importance of gas component parameters, and comprehensively evaluate the importance of each gas component parameter to the gas carbon emission factor;

[0056] S5. Establish an optimal prediction model for gas carbon emission factors: Based on the numerical values ​​of the comprehensive importance evaluation indicators of each gas component parameter, a random forest combined with a backward sequence selection algorithm is used to establish an optimal prediction model for gas carbon emission factors;

[0057] S6. Data verification: The third-party verification agency inputs the various gas component parameters of the gas-fired power generation unit reported by the enterprise into the gas carbon emission factor prediction model established in step S5 to obtain the model prediction value, and then calculates the absolute value of the error between the model prediction value and the gas carbon emission factor data reported by the enterprise, and compares the absolute value of the error with the preset threshold: if the absolute value of the error is less than or equal to the preset threshold, the data is judged to be reasonable; if the absolute value of the error is greater than the preset threshold, the data is judged to be questionable and further verification of the various gas component parameter data of the gas-fired power generation unit is required. Before the third-party verification agency inputs the various gas component parameters of the gas-fired power generation unit reported by the enterprise into the gas carbon emission factor prediction model, the input gas component parameter data should be preprocessed as in the subsequent step S5.1.

[0058] In some embodiments, the sensitivity coefficient r between each gas component parameter and the gas carbon emission factor is calculated in S2 according to the following formula:

[0059] ;

[0060] Where r i is the sensitivity coefficient of the i-th gas component parameter, I ij is the i-th gas component parameter of the j-th sample, O j is the carbon emission factor of the gas of the jth sample, n is the number of gas samples, and m is the number of gas component parameters of each gas sample (for example, the sensitivity coefficient of the first gas component parameter is based on the first gas component parameter I of n samples). 11 , I 12 , I 13 , I 14 , I 15 .....I 1n And the gas carbon emission factors O1, O2, O3, O4, ....O corresponding to n samples n The sensitivity coefficient r ranges from 0 to 1. The closer the value of the sensitivity coefficient of the gas component parameter is to 1, the stronger the correlation between the gas component parameter and the gas carbon emission factor is, that is, the change of the component parameter will significantly affect the result of the gas carbon emission factor; a value close to 0 means no linear correlation.

[0061] In some embodiments, in S3, a random forest algorithm is used to calculate the characteristic importance M of each gas component parameter. The larger the characteristic importance M of a gas component parameter, the greater the impact of the gas component parameter on the performance of the gas carbon emission factor prediction model. For each gas component parameter (e.g., the i-th gas component parameter), the original prediction error e is first calculated based on the trained random forest model (the training process of the random forest model is the same as the method of subsequent steps S5.1 to S5.5) using the out-of-bag samples corresponding to each decision tree (i.e., sample data that was not selected by the decision tree during training, equivalent to the test set). t (i.e., the mean absolute error between the original gas carbon emission factor model prediction value and the actual value of the gas carbon emission factor. The original gas carbon emission factor model prediction value is the model prediction value obtained by the random forest model before the gas component parameters are randomly shuffled). Then, after randomly shuffling the values ​​of the gas component parameter i in the out-of-bag samples, the prediction error is recalculated. ; Finally, calculate the difference in prediction error before and after randomly disrupting the gas component parameters to obtain the characteristic importance of the gas component parameters. That is, if a gas component parameter is crucial to the prediction of the gas carbon emission factor model, disrupting its value will cause the model's prediction error to increase significantly, thereby increasing the M value; conversely, the error changes slightly after irrelevant features are disrupted, and the M value approaches zero, that is, the larger the value of the characteristic importance M, the greater the impact of the gas component parameter on the performance of the gas carbon emission factor prediction model. Finally, by comparing the characteristic importance of each gas component parameter, its contribution to the gas carbon emission factor prediction model can be quantified. The characteristic importance M is calculated as follows:

[0062] ;

[0063] Where, is the feature importance of the i-th gas component parameter, i represents the i-th gas component parameter, t represents the t-th tree, is the tree containing the gas component parameter i, is the number of trees using gas component parameter i, is the prediction error calculated by randomly disrupting the sample data of the i-th gas component parameters, The prediction error of the current tree is calculated using out-of-bag samples that did not participate in the training of tree t.

[0064] In some embodiments, in S4, when establishing a comprehensive evaluation index for the importance of gas component parameters, the weight of the sensitivity coefficient of the gas component parameter is the same as the weight of the feature importance. Specifically, based on the sensitivity coefficient and the feature importance obtained by random forest, the advantages of the two evaluation methods are combined to establish a comprehensive evaluation index CI for the importance of gas component parameters, thereby comprehensively evaluating the importance of gas component parameters to the gas carbon emission factor. Specifically, the fusion is performed according to the following formula:

[0065] ;

[0066] Where, is the comprehensive importance evaluation index of the i-th gas component parameter, r i is the sensitivity coefficient, is the feature importance, a and b are weight coefficients, both of which are selected as 0.5 here.

[0067] The sensitivity coefficient r reflects the statistical correlation between gas component parameters and carbon emission factors, while the feature importance M reflects the contribution of gas component parameters to the prediction model. This dual verification mechanism not only captures the trend correlation of gas component parameter changes (via the sensitivity coefficient r) but also assesses the actual predictive value of gas component parameters in the machine learning model (via the feature importance M). The larger the comprehensive importance evaluation index of a gas component parameter, the more important it is to the gas carbon emission factor.

[0068] Preferably, in said S5, the method for establishing a fuel gas carbon emission factor prediction model includes:

[0069] S5.1. Data preprocessing: Based on the gas component parameters in step S1 (methane (CH4), ethane (C2H6), propane (C3H8), isobutane (i-C4H 10 ), n-butane (n-C4H 10 ), isopentane (i-C5H 12 ), n-pentane (n-C5H 12 A set of gas samples with known gas composition (e.g., carbon dioxide (CO2), nitrogen (N2), and lower calorific value (NCV)) are used as training samples. A full-variable random forest regression model of the gas carbon emission factor is established using the random forest algorithm. The initial data set of the gas samples is preprocessed to obtain a preprocessed data set. The data in the initial data set is the original measurement data of the gas samples. The original measurement data includes the gas component parameters, elemental carbon content, and lower calorific value data of each gas sample. The original measurement data is used as the original data of the training samples. The gas component data of the gas samples used for training must be standard gas data, or gas data tested by a CMA-accredited or CNAS-certified testing agency / laboratory.

[0070] S5.2. Self-service sampling: After preprocessing the training sample, the dataset N={(x j ,y j )}(j=1, 2, ..., n) performs bootstrap sampling, that is, randomly sampling with replacement in the original preprocessed dataset N, extracting s samples each time to form a new training set. This process is repeated until the size of each new training set is the same as the preprocessed dataset:

[0071] ;

[0072] Where, is the b-th sub-dataset, is the sample index set of the b-th sub-dataset, x j is the independent variable of the jth sample, which refers to the gas component parameter, y jis the dependent variable of the jth sample, which refers to the gas carbon emission factor;

[0073] S5.3. Parameter Selection: Use grid search and cross-validation to determine the optimal parameters for random forest modeling. Then, based on the selected optimal parameters, determine the number of decision trees and the minimum number of leaves, and construct the decision tree.

[0074] S5.4. Decision Tree Construction: Build a decision tree using each new training set obtained through the self-sampling in step S5.2. During the decision tree construction process, for each node in the tree, instead of considering all features for optimal segmentation, a portion of the gas component parameter features is randomly selected from all gas component parameter features, and then the optimal segmentation feature is selected from these randomly selected features:

[0075]

[0076] Where, To finally determine the optimal segmentation feature, argmin is a mathematical operator that represents selecting the parameter that minimizes the objective function value from a given set. j is the feature index, S is the randomly selected feature subset, and n is the number of samples in the current node. is the true value of the gas carbon emission factor of the i-th sample, is the predicted value of the gas carbon emission factor of the i-th sample after splitting according to feature j;

[0077] S5.5. Integrated prediction: After all decision trees are trained, a gas carbon emission factor prediction model is formed;

[0078] After the gas carbon emission factor prediction model is formed, the gas carbon emission factor prediction model uses each decision tree to predict the newly input sample data, and then averages all the prediction results to obtain the final model prediction value :

[0079]

[0080] Where, is the final predicted value of the input sample x, T is the number of decision trees, is the predicted value of the t-th tree.

[0081] In some embodiments, establishing the gas carbon emission factor prediction model in S5 further includes: using a backward sequence selection algorithm combined with a full-variable random forest regression model, according to the descending order of the comprehensive importance evaluation index of the gas component parameters, successively eliminating the single gas component parameters with the lowest current comprehensive importance evaluation index value (such as first eliminating n-pentane with the smallest CI value), and reconstructing a new gas carbon emission factor prediction model.

[0082] In some embodiments, S5 further includes using the mean absolute percentage error (MAPE) as a performance evaluation indicator of the gas carbon emission factor prediction model:

[0083]

[0084] In the formula, yi, are the reference value and model prediction value of the gas carbon emission factor, n represents the number of gas samples, and i is the number of independent variables.

[0085] In the subsequent step S6, before the third-party verification agency inputs the various gas component parameters of the gas-fired power generation unit reported by the enterprise into the gas carbon emission factor prediction model, the input data should be preprocessed as in step S5.1.

[0086] After each removal of a single gas component parameter with the lowest value for the current comprehensive importance evaluation index, the rationality of feature removal is evaluated by calculating the change in the MAPE (Massive Performance Evaluation) of the new gas carbon emission factor prediction model performance indicator. The removal threshold is set to a relative increase in MAPE of no more than 10% (i.e., if the increase in the new model's MAPE is ≤10%, performance is considered to have not significantly degraded, and the parameter is retained; otherwise, the removed parameter is restored). Through this iterative backward screening, redundant gas component parameters are gradually removed (if the increase in the new model's MAPE is ≤10% after removing a gas component parameter, the removed gas component parameter is considered redundant) until further removal of any gas component parameter results in a MAPE increase exceeding the 10% threshold. Finally, based on the prediction model performance evaluation index and the number of gas component parameters, the gas component parameters most relevant to the gas carbon emission factor are comprehensively selected. The optimal gas carbon emission factor prediction model is constructed using these finally selected gas component parameters. Redundant gas component parameters are those that have little impact on the performance indicators of the gas carbon emission factor prediction model. By reducing irrelevant features, the generalization error of the model can be reduced and the predictive ability of the model can be improved.

[0087] In some embodiments, step S5.1 includes: normalizing the input and output data using a minimum-maximum normalization method to map the data to the interval [0, 1], specifically by calculating:

[0088]

[0089]

[0090] Where, X norm is the normalized value of the input data, X is the original input data value, and X min is the minimum value of the original input data, Xmax is the maximum value of the original input data; Y norm is the normalized value of the output data, Y is the original output data value, and Y min is the minimum value of the original output data, Y max is the maximum value of the original output data;

[0091] Step S5.3 includes: optimizing the core parameters of the number of decision trees (trees) and the minimum number of leaves (min_leaf); establishing a complete parameter search space by setting candidate ranges for the number of decision trees and the minimum number of leaves; calculating the predictive performance of each parameter combination using a 5-fold cross-validation strategy and mean-squared error (MSE) as the evaluation metric, and ultimately selecting the optimal parameter combination that minimizes the MSE. After parameter optimization is complete, the optimal parameters are fixed as the modeling configuration.

[0092] In some embodiments, a prediction model performance evaluation is further included to evaluate the prediction performance of the gas carbon emission factor prediction model established above. A set of test set samples is used to evaluate the prediction performance of the prediction model established in step S5. Generally, a smaller RMSE indicates a higher prediction accuracy of the model.

[0093]

[0094] In the formula, yi, are the reference value and model prediction value of the gas carbon emission factor, n represents the number of gas samples, and i is the number of independent variables.

[0095] The specific implementation case of Example 1 is as follows:

[0096] (I) Sample collection and processing: This implementation case collected a component dataset of gas samples used in the combustion of 53 gas-fired generator sets. The gas component dataset includes methane (CH4), ethane (C2H6), propane (C3H8), isobutane (i-C4H 10 ), n-butane (n-C4H 10 ), isopentane (i-C5H 12 ), n-pentane (n-C5H 12 ), carbon dioxide (CO2), nitrogen (N2), elemental carbon content (C), and lower calorific value (NCV) are all data obtained by testing agencies certified by CNAS.

[0097] Based on the elemental carbon content and lower calorific value of the above gas component dataset, the combustion carbon emission factors of the 53 gas samples were calculated using the following formula:

[0098]

[0099] Where, CEF is the carbon emission factor of the gas, measured in tCO2 / GJ; C is the elemental carbon content of the gas, measured in tC / t; NCV is the lower calorific value of the gas, measured in GJ / t; OF is the oxidation rate of the gas, measured in %. The carbon oxidation rate of the gas specified in the current accounting guidelines is 99%.

[0100] (II) Sensitivity analysis: A sensitivity analysis was conducted on the gas carbon emission factor and gas component parameters. The sensitivity coefficients of the gas carbon emission factor and each parameter (expressed as cosine similarity) were calculated according to the following formula. The closer the value of the sensitivity coefficient of the gas component parameter is to 1, the stronger the correlation between the gas component parameter and the gas carbon emission factor is, that is, the change of the component parameter will significantly affect the result of the gas carbon emission factor; a value close to 0 means there is no linear correlation:

[0101]

[0102] Where r i is the sensitivity coefficient of the i-th gas component parameter, I ij is the i-th gas component parameter of the j-th sample, O j is the gas carbon emission factor of the jth sample, n is the number of gas samples, and m is the number of gas component parameters of each gas sample.

[0103] The calculation results are as follows Figure 2 As shown in the figure, it can be seen that the sensitivity coefficients of the gas component parameters are ranked as follows: methane > lower calorific value > ethane > nitrogen > propane > isobutane > n-butane > carbon dioxide > isopentane > n-pentane.

[0104] (III) Feature Importance Analysis: Using the random forest algorithm, the feature importance of each gas component parameter is calculated according to the following formula. The larger the value, the greater the impact of the gas component parameter on the performance of the gas carbon emission factor prediction model:

[0105]

[0106] Where, is the feature importance of the i-th gas component parameter, i represents the i-th gas component parameter, t represents the t-th tree, is the tree containing the gas component parameter i, is the number of trees using gas component parameter i, is the prediction error calculated by randomly disrupting the sample data of the i-th gas component parameters, The prediction error of the current tree is calculated using out-of-bag samples that did not participate in the training of tree t.

[0107] The calculation results are as follows Figure 3As shown in the figure, it can be seen that the order of the importance of the characteristic parameters of the gas components is: carbon dioxide > lower calorific value > methane > isobutane > n-butane > isopentane > propane > n-pentane > ethane > nitrogen.

[0108] (IV) Based on the above sensitivity coefficients and characteristic importance, the comprehensive importance evaluation index of the gas component parameters is calculated according to the following formula to comprehensively evaluate the importance of the gas component parameters to the gas carbon emission factor. The larger the comprehensive importance evaluation index of the gas component parameters, the more important the gas component parameters are to the gas carbon emission factor:

[0109]

[0110] Where, is the comprehensive importance evaluation index of the i-th gas component parameter, ri is the sensitivity coefficient, is the feature importance, a and b are weight coefficients, both of which are selected as 0.5. The results are shown in Table 1 below:

[0111] Table 1 Calculation table of comprehensive importance evaluation index of gas component parameters

[0112] Serial number parameter Sensitivity coefficient Feature Importance Comprehensive importance evaluation index 1 methane 0.9996 0.4867 0.7431 2 Ethane 0.9845 0.1808 0.5827 3 Propane 0.8596 0.2635 0.5616 4 Isobutane 0.8110 0.4188 0.6149 5 n-butane 0.7636 0.4090 0.5863 6 Isopentane 0.7047 0.2747 0.4897 7 n-pentane 0.6467 0.2617 0.4542 8 carbon dioxide 0.7128 0.8190 0.7659 9 Nitrogen 0.9170 0.0739 0.4954 10 Low calorific value 0.9990 0.6082 0.8036

[0113] It can be seen from the table that the order of the comprehensive importance evaluation indicators of gas component parameters is: lower calorific value > carbon dioxide > methane > isobutane > n-butane > ethane > propane > nitrogen > isopentane > n-pentane.

[0114] (V) Establish a prediction model for gas carbon emission factors: First, standardize the input and output data using the minimum-maximum normalization method according to the following formula and map the data to the [0, 1] interval:

[0115]

[0116]

[0117] Where Xnorm is the normalized value of the input data, X is the original input data value, Xmin is the minimum value of the original input data, and Xmax is the maximum value of the original input data; Ynorm is the normalized value of the output data, Y is the original output data value, Ymin is the minimum value of the original output data, and Ymax is the maximum value of the original output data.

[0118] Next, we used a grid search method and cross-validation to determine the optimal parameters for random forest modeling, focusing on tuning core parameters such as the number of decision trees (trees) and the minimum number of leaves (min_leaf). By setting the candidate range for the number of decision trees to [50, 100, 150, 200] and the candidate set for the minimum number of leaves to [1, 3, 5, 10], we established a complete parameter search space. Using a 5-fold cross-validation strategy, we ultimately determined the optimal parameters: 100 decision trees and 3 minimum leaves.

[0119] Based on 10 gas component parameters (methane (CH4), ethane (C2H6), propane (C3H8), isobutane (i-C4H 10 ), n-butane (n-C4H 10 ), isopentane (i-C5H 12 ), n-pentane (n-C5H 12 ), carbon dioxide (CO2) content, as well as nitrogen (N2) and lower heating value (NCV)) to establish a full-variable random forest regression model for gas carbon emission factors as a benchmark. Then, using the Sequential Backward Selection (SBS) algorithm combined with the Random Forest algorithm, the gas component parameters are sorted in descending order of their comprehensive importance evaluation index, and the individual parameters with the lowest current comprehensive importance evaluation index are successively removed (for example, n-pentane with the smallest CI value is initially removed), and a new prediction model is rebuilt. After each feature removal, the rationality of the feature removal is evaluated by calculating the change in the model performance indicator MAPE (mean absolute percentage error). The removal threshold is set to a relative increase in MAPE of no more than 10% (i.e., if the MAPE increase of the new model is ≤10%, the performance is considered to have not significantly decreased, and the removal is retained; otherwise, the removed parameter is restored). Through this iterative backward screening, redundant variables are gradually removed until further parameter removal results in a MAPE increase exceeding the 10% threshold. Ultimately, a feature combination is obtained that strikes the optimal balance between forecast accuracy (minimizing MAPE) and model simplicity (minimizing the number of variables). As shown in Table 2, this process clearly demonstrates the impact of parameter removal at each stage on model performance.

[0120] Table 2 Model variable selection

[0121] Number of parameters parameter MAPE(%) 10 <h2 style=";text-align:left;direction:ltr"><![CDATA[NCV、CO2、CH4、i-C4H <h2 style=";text-align:left;direction:ltr"> 10 <h2 style=";text-align:left;direction:ltr"> n-C4H<h2 style=";text-align:left;direction:ltr"> 10 <h2 style=";text-align:left;direction:ltr"> 、C2H6、C3H8、N2、i-C5H<h2 style=";text-align:left;direction:ltr"> 12 <h2 style=";text-align:left;direction:ltr"> n-C5H<h2 style=";text-align:left;direction:ltr"> 12 <h2 style=";text-align:left;direction:ltr"> ]]><h2 style=";text-align:left;direction:ltr"> 0.1338 9 <![CDATA[NCV、CO2、CH4、i-C4H 10 、n-C4H 10 、C2H6、C3H8、N2、i-C5H 12 ]]> 0.1389 8 <![CDATA[NCV、CO2、CH4、i-C4H 10 、n-C4H 10 、C2H6、C3H8、N2]]> 0.1463 7 <![CDATA[NCV、CO2、CH4、i-C4H 10 、n-C4H 10 、C2H6、C3H8]]> 0.1463 6 <![CDATA[NCV、CO2、CH4、i-C4H 10 、n-C4H 10 、C2H6]]> 0.1472 5 <![CDATA[NCV、CO2、CH4、i-C4H 10 、n-C4H 10 ]]> 0.1421 4 <![CDATA[NCV、CO2、CH4、i-C4H 10 ]]> 0.1386 3 <![CDATA[NCV、CO2、CH4]]> 0.1532 2 <![CDATA[NCV、CO2]]> 0.1645 1 NCV 0.2588

[0122] According to the model performance indicators and the number of parameters, the four gas component parameters that are most relevant to the gas carbon emission factor are comprehensively selected: lower calorific value (NCV), carbon dioxide (CO2), methane (CH4), isobutane (i-C4H 10 ), and the four parameters are used to construct a prediction model for the optimal gas carbon emission factor.

[0123] (VI) Evaluation of prediction model performance: 13 gas samples were used as test set samples to evaluate the prediction performance of the above-built model. The gas composition parameters of the test set samples (methane (CH4), lower calorific value (NCV), ethane (C2H6), nitrogen (N2), propane (C3H8), isobutane (i-C4H 10 )) is substituted into the above-established gas carbon emission factor prediction model to obtain the predicted value of the gas carbon emission factor, which is compared with the gas carbon emission factor calculated by actual sample measurement. The results are shown in Table 3 below:

[0124] Table 3. Absolute value calculation table of test set sample error

[0125] Test sample serial number <![CDATA[Measured calculated value (t CO2 / GJ)]]> <![CDATA[Model predicted value (t CO2 / GJ)]]> <![CDATA[Absolute error (t CO2 / GJ)]]> 1 55.504 55.785 0.281 2 55.798 55.901 0.103 3 55.852 55.927 0.075 4 55.952 55.915 0.037 5 56.061 55.967 0.094 6 56.147 55.996 0.151 7 56.369 56.637 0.268 8 56.698 56.224 0.474 9 56.935 56.913 0.022 10 57.043 57.017 0.026 11 57.111 57.070 0.041 12 58.082 58.113 0.031 13 58.266 58.192 0.074

[0126] As shown in Table 3 above, the root mean square error (RMSE) for the test set samples is 0.18 tCO₂ / GJ, indicating good predictive performance of the model. A 2RMSE (0.36 tCO₂ / GJ) is used as the reasonableness threshold. As shown in Table 3, with the exception of test sample 8, the absolute values ​​of the errors between the model-predicted and measured gas carbon emission factor values ​​for the other 12 gas test samples are within the threshold. This indicates that the gas carbon emission factor data for test sample 8 is questionable, requiring further verification of the relevant gas component parameters for this sample. The gas carbon emission factor data for the other 12 gas test samples are all within a reasonable range.

[0127] Example 2

[0128] This embodiment discloses a system for predicting and verifying the carbon emission factor of a gas-fired generator set. The system executes the above method and includes:

[0129] Data collection and processing module, importance comprehensive evaluation index generation module, prediction model establishment module, data verification module;

[0130] The data collection and processing module is used to collect and process samples, collect a group of gas samples with known component data sets, obtain a gas component data set of the group of gas samples, and the gas component data set includes gas component parameters, elemental carbon content, and lower calorific value data of each gas sample; and calculate the gas carbon emission factor of each gas sample according to the elemental carbon content and lower calorific value data of each gas sample in the gas component data set;

[0131] The importance comprehensive evaluation index generation module is used to perform sensitivity analysis and characteristic importance analysis on each gas component parameter in the group of gas samples, and integrate the sensitivity coefficient and characteristic importance of each gas component parameter to establish a comprehensive evaluation index of the importance of the gas component parameter, and comprehensively evaluate the importance of each gas component parameter to the gas carbon emission factor; wherein: sensitivity analysis includes: analyzing the sensitivity between the gas carbon emission factor and each gas component parameter in the gas component data set, and ranking the size of the sensitivity coefficient of each gas component parameter to analyze the influence of each gas component parameter on the gas carbon emission factor; characteristic importance analysis includes: calculating the characteristic importance of each gas component parameter, and ranking the size of the characteristic importance of each gas component parameter. By comparing the characteristic importance of each gas component parameter, its contribution to the gas carbon emission factor prediction model can be quantified;

[0132] The prediction model establishment module is used to establish a gas carbon emission factor prediction model based on the numerical value of the comprehensive importance evaluation index of each gas component parameter using a random forest combined with a backward sequence selection algorithm;

[0133] The data verification module is used to input the various gas component parameters of the gas generator set reported by the enterprise into the gas carbon emission factor prediction model to obtain the model prediction value, calculate the absolute value of the error between the model prediction value and the gas carbon emission factor data reported by the enterprise, and determine whether the data reported by the enterprise is reasonable.

[0134] The above is a detailed description of an embodiment of a method and system for predicting and verifying the carbon emission factor of a gas-fired power generation unit provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified in several ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting and verifying the carbon emission factor of a gas-fired power generation unit, characterized in that: The steps include: S1. Collect a set of gas samples with known composition data sets and obtain a gas composition data set for the set of gas samples, the gas composition data set including gas component parameters, elemental carbon content, and lower calorific value data for each gas sample; Calculate the carbon emission factor of each gas sample according to the elemental carbon content and lower calorific value data of each gas sample in the gas component data set; S2. Analyze the sensitivity between each gas carbon emission factor in the group of gas samples and each gas component parameter in the gas component dataset to obtain the sensitivity coefficient between each gas component parameter and the gas carbon emission factor, and rank the sensitivity coefficients of each gas component parameter; S3. Calculate the characteristic importance of each gas component parameter and rank the importance of each gas component parameter; S4. Combine the sensitivity coefficients and characteristic importance of each gas component parameter to establish a comprehensive evaluation index of the importance of the gas component parameters; S5. Based on the numerical values ​​of the comprehensive importance evaluation indicators of each gas component parameter, a gas carbon emission factor prediction model is established, including the following steps: S5.

1. Data Preprocessing: Based on the set of gas samples with known gas composition parameters from step S1, a full-variable random forest regression model for the gas carbon emission factor is established using the random forest algorithm. The initial dataset of gas samples is preprocessed to obtain a preprocessed dataset. The data in the initial dataset consists of the raw measurement data of the gas samples, including the gas composition parameters, elemental carbon content, and lower heating value data for each gas sample. This raw measurement data serves as the raw data for the training samples. S5.

2. Bootstrap sampling: Bootstrap sampling is performed on the preprocessed dataset from the training set, with n samples drawn each time to form a new training set. This process is repeated until the size of each new training set is the same as the preprocessed dataset. S5.

3. Parameter Selection: Use grid search and cross-validation to determine the optimal parameters for random forest modeling. Then, based on the selected optimal parameters, determine the number of decision trees and the minimum number of leaves, and construct the decision tree. S5.

4. Decision Tree Construction: Construct a decision tree using each new training set obtained through the self-sampling in step S5.

2. During the decision tree construction process, for each node in the tree, randomly select a portion of the gas component parameter features from all gas component parameter features, and then select the optimal segmentation feature from these randomly selected features. S5.

5. After all decision trees are trained, an initial gas carbon emission factor prediction model is formed; S5.

6. Use the backward sequence selection algorithm in conjunction with the initial gas carbon emission factor prediction model. Arrange the gas component parameters in descending order of their comprehensive importance evaluation indexes, and sequentially eliminate the individual gas component parameters with the lowest current comprehensive importance evaluation index values. Reconstruct a new gas carbon emission factor prediction model to form the final gas carbon emission factor prediction model. After the final gas carbon emission factor prediction model is formed, the final gas carbon emission factor prediction model uses each decision tree to predict the newly input sample data, and then averages all the prediction results to obtain the final model prediction value; S6. Input the various gas component parameters of the gas-fired power generation unit reported by the enterprise into the final gas carbon emission factor prediction model to obtain the model prediction value, calculate the absolute value of the error between the model prediction value and the gas carbon emission factor data reported by the enterprise, and determine whether the data reported by the enterprise is reasonable.

2. The method for predicting and verifying the combustion carbon emission factor of a gas-fired generator set according to claim 1, characterized in that: In S2, the sensitivity coefficient r between each gas component parameter and the gas carbon emission factor is calculated according to the following formula: ； Where r i is the sensitivity coefficient of the i-th gas component parameter, I ij is the i-th gas component parameter of the j-th sample, O j is the gas carbon emission factor of the jth sample, n is the number of gas samples, and m is the number of gas component parameters of each gas sample. The sensitivity coefficient r ranges from 0 < r < 1. The closer the value of the sensitivity coefficient of the gas component parameter is to 1, the stronger the correlation between the gas component parameter and the gas carbon emission factor.

3. The method for predicting and verifying the combustion carbon emission factor of a gas-fired generator set according to claim 1, characterized in that: In S3, the random forest algorithm is used to calculate the feature importance M of each gas component parameter. For each gas component parameter, the original prediction error e is first calculated based on the trained random forest model using the out-of-bag samples corresponding to each decision tree. t , the original prediction error e t is the mean absolute error between the original gas carbon emission factor model prediction value and the actual value of the gas carbon emission factor; then the values ​​of the gas component parameters in the out-of-bag samples are randomly disrupted and the prediction error is recalculated Finally, the difference in prediction error before and after randomly disrupting the gas component parameters is calculated to obtain the characteristic importance M of the gas component parameters; The larger the characteristic importance M value of a certain gas component parameter is, the greater the impact of the gas component parameter on the performance of the gas carbon emission factor prediction model is.

4. The method for predicting and verifying the combustion carbon emission factor of a gas-fired power generation unit according to claim 1, characterized in that: In the above S4, in the process of establishing the comprehensive evaluation index of the importance of the gas component parameters, the weight of the sensitivity coefficient of the gas component parameters is the same as the weight of the characteristic importance.

5. The method for predicting and verifying the combustion carbon emission factor of a gas-fired power generation unit according to claim 1, characterized in that: Said S5 also includes using the mean absolute percentage error (MAPE) as a performance evaluation indicator of the gas carbon emission factor prediction model: Each time after removing a single gas component parameter with the lowest value of the current comprehensive importance evaluation index, the rationality of feature removal is evaluated by calculating the change range of the new gas carbon emission factor prediction model performance indicator MAPE. The removal threshold is set, and redundant gas component parameters are gradually removed until continuing to remove any gas component parameter will cause the MAPE increase to exceed the removal threshold.

6. The method for predicting and verifying the combustion carbon emission factor of a gas-fired power generation unit according to claim 1, characterized in that: S5.1 includes: normalizing the input and output data using a minimum-maximum normalization method to map the data to the interval [0, 1]; The S5.3 includes: tuning the core parameters of the number of decision trees and the minimum number of leaves, establishing a complete parameter search space by setting the candidate range of the number of decision trees and the minimum number of leaves; based on a 5-fold cross-validation strategy, using the mean square error as the evaluation indicator, calculating the prediction performance of the gas component parameter combination, and finally selecting the optimal parameter combination that minimizes the MSE; after completing the optimal parameter selection, fixing the optimal parameters as the modeling configuration.

7. The method for predicting and verifying the combustion carbon emission factor of a gas-fired power generation unit according to claim 1, characterized in that: It also includes prediction model performance evaluation: using a set of test set samples to evaluate the prediction performance of the prediction model built in step S5.

8. A system for predicting and verifying carbon emission factors of gas-fired power generation units, characterized in that: The system executes the method according to any one of claims 1 to 7, and the system includes: Data collection and processing module, importance comprehensive evaluation index generation module, prediction model establishment module, data verification module; The data collection and processing module is used to collect and process samples, collect a group of gas samples with known component data sets, obtain a gas component data set of the group of gas samples, and the gas component data set includes gas component parameters, elemental carbon content, and lower calorific value data of each gas sample; and calculate the gas carbon emission factor of each gas sample according to the elemental carbon content and lower calorific value data of each gas sample in the gas component data set; The importance comprehensive evaluation index generation module is used to perform sensitivity analysis and characteristic importance analysis on each gas component parameter in the group of gas samples, and integrate the sensitivity coefficient and characteristic importance of each gas component parameter to establish a comprehensive evaluation index of the importance of the gas component parameter, and comprehensively evaluate the importance of each gas component parameter to the gas carbon emission factor; The prediction model establishment module is used to establish a gas carbon emission factor prediction model based on the numerical value of the comprehensive importance evaluation index of each gas component parameter using a random forest combined with a backward sequence selection algorithm; The data verification module is used to input the various gas component parameters of the gas generator set reported by the enterprise into the gas carbon emission factor prediction model to obtain the model prediction value, calculate the absolute value of the error between the model prediction value and the gas carbon emission factor data reported by the enterprise, and determine whether the data reported by the enterprise is reasonable.

Images