Park-level monthly carbon emission monitoring method, system, equipment and medium
By employing Pearson correlation analysis, conditional generative adversarial networks, and gradient boosting decision tree algorithms, a monthly carbon emission monitoring method for industrial parks was constructed. This method addresses the issues of untimely, inaccurate, and inconvenient carbon emission monitoring at the industrial park level, and enables high-precision monthly carbon emission prediction and dynamic management.
Patent Information
- Application Number
- CN202511575975.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-31
- Publication Date
- 2026-02-17
AI Technical Summary
Existing carbon emission monitoring methods suffer from problems such as untimely, inaccurate, and inconvenient monitoring at the park level, making it difficult to achieve dynamic and refined management. Furthermore, the lack of high-frequency economic data makes model training and validation difficult.
Pearson correlation analysis was used to screen key factors, conditional generative adversarial networks were combined to generate monthly economic data, random forest algorithm was used to evaluate feature importance, and gradient boosting decision tree algorithm was used to build a carbon emission monitoring model, thereby achieving data frequency dimensionality upgrade and information reconstruction from annual to monthly.
It has achieved high-precision monthly carbon emission forecasting, solved the problem of missing high-frequency data, improved the timeliness and accuracy of monitoring, and provided reliable technical support for carbon management in the park.
Smart Images
Figure CN121543867A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of carbon emission monitoring technology, and in particular to a method, system, equipment and medium for monthly carbon emission monitoring at the park level. Background Technology
[0002] Traditional carbon emission monitoring methods mainly include emission factor-based calculation methods and material balance methods. The former uses static emission factors combined with activity data for linear calculation, but it is difficult to reflect dynamic emission characteristics and cannot handle the spatiotemporal fluctuations of indirect emissions, resulting in poor data granularity and timeliness. The latter is suitable for carbon flow tracking in process industrial parks, but it requires complete material data, has high implementation costs, and is sensitive to data gaps. These methods have significant shortcomings in dynamic monitoring and refined management. Data-driven methods provide a new technical path for carbon emission monitoring. Remote sensing technology (such as satellite inversion) can achieve regional emission monitoring, but its low resolution makes it difficult to meet the needs of park-scale monitoring. IoT sensor networks improve monitoring accuracy by collecting data in real time, but high-precision sensors are expensive, and sparse data points can lead to interpolation errors. Among machine learning methods, models such as random forests / XGBoost can handle nonlinear relationships, but they are prone to overfitting on small sample data. Time series models (such as ARIMA and LSTM) are good at trend prediction, but they have high requirements for data uniformity and are difficult to adapt to sudden operating conditions. Overall, existing methods have not yet achieved a balance between dynamism, accuracy, and cost-effectiveness, and there is an urgent need to integrate multi-source data with novel algorithms to overcome technical bottlenecks.
[0003] Currently, methods for calculating carbon emissions by utilizing energy consumption data obtained from publicly available government websites such as statistical yearbooks and combining it with carbon emission factors for various energy consumption categories lack real-time performance. This makes it difficult to monitor actual carbon emissions in a timely manner and provides limited assistance in guiding the formulation of carbon reduction policies for industrial parks. To achieve timely monitoring of carbon emissions across all provinces and cities nationwide, many provinces and cities have constructed carbon emission calculation models. However, due to the availability of basic data involved in these models, carbon emission monitoring is mostly limited to quarterly or annual dimensions, failing to achieve higher-frequency monitoring. Furthermore, in terms of regional coverage, existing research primarily focuses on the national and provincial levels for ease of data acquisition, with a relative lack of research on carbon emission monitoring at the industrial park level. This results in insufficient precision and specificity in calculating carbon emissions at the industrial park level. Summary of the Invention
[0004] In view of the aforementioned existing problems, the present invention is proposed.
[0005] Therefore, this invention provides a method, system, equipment, and medium for monthly carbon emission monitoring at the park level to solve the problems of untimely, inaccurate, and inconvenient monitoring of carbon emissions at the park level in existing carbon emission monitoring methods.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for monitoring monthly carbon emissions at the park level, comprising: Obtain data on the park's annual carbon emissions and candidate influencing factors; Using the Pearson correlation analysis algorithm, the correlation between annual carbon emissions and candidate influencing factor data is calculated and compared to obtain key factors that are highly correlated with carbon emissions. By using a conditional generative adversarial network, with the park's annual economic indicator data as conditional constraint information and combined with random noise, a simulated monthly economic indicator data sequence is generated. The random forest algorithm is used to assess the feature importance of various indicators related to annual carbon emissions. Based on the importance scores obtained from the assessment, the annual carbon emissions are weighted and split into monthly data to obtain the historical monthly carbon emission data. The model dataset is formed by integrating monthly economic indicator data series, historical monthly carbon emission data, and key factors highly correlated with carbon emissions. The gradient boosting decision tree algorithm is used to extract features from the model dataset. The extracted key features are used as input to the model to construct a monthly carbon emission monitoring model at the park level and output the monthly carbon emission monitoring results at the park level.
[0007] As a preferred embodiment of the park-level monthly carbon emission monitoring method described in this invention, the key factors highly correlated with carbon emissions include: The annual carbon emissions are combined with the annual data of each candidate influencing factor to form a data set; For each data combination, calculate the corresponding Pearson correlation coefficient; Pre-set a threshold for determining the absolute value of correlation; The absolute value of the Pearson correlation coefficient corresponding to each data combination is compared with the judgment threshold. Candidate influencing factors whose absolute value of the Pearson correlation coefficient is not lower than the determination threshold are identified as key factors.
[0008] As a preferred embodiment of the park-level monthly carbon emission monitoring method described in this invention, the generation of the simulated monthly economic indicator data sequence includes: Using the park's annual economic indicators as input, a conditional generative adversarial network containing a generator and a discriminator is constructed. In each iteration, the generator parameters are fixed, and the discriminator is trained by inputting real monthly economic indicator data and conditional input. In each iteration, the discriminator parameters are fixed, and random noise and conditional labels are input to train the generator; The training process for the discriminator and generator is performed alternately to continuously optimize the parameters of the discriminator and generator; Using the trained generator, combined with new annual economic indicator data as conditional input and random noise, simulated monthly economic indicator data sequences are generated.
[0009] The beneficial effect of this preferred technical solution is that it generates high-quality monthly economic data sequences by using conditional generative adversarial networks with annual economic indicators as conditions, effectively solving the problem of missing high-frequency economic data in monthly carbon emission monitoring and providing reliable input for subsequent models.
[0010] As a preferred embodiment of the park-level monthly carbon emission monitoring method described in this invention, the step of obtaining historical monthly carbon emission data includes: Obtain annual carbon emission data at the park level, as well as related annual business activity indicator datasets; The annual business activity indicator dataset was processed using the random forest algorithm to calculate the importance score of each business activity indicator in the process of predicting annual carbon emissions. The importance score of each business activity indicator will be used as the weight of the current business activity indicator. The monthly data for each business activity indicator is multiplied by its corresponding importance score to obtain the weight. By summing the weighted monthly data of all business activity metrics, we can obtain the historical carbon emission data for each month.
[0011] The beneficial effect of this preferred technical solution is that it evaluates the impact weight of business activity indicators on annual carbon emissions through the random forest algorithm, and reasonably decomposes the annual carbon emissions into monthly dimensions based on the weight, effectively solving the problem of missing historical monthly carbon emission data, and providing reliable label data for building a high-precision monthly monitoring model.
[0012] As a preferred embodiment of the park-level monthly carbon emission monitoring method described in this invention, the output of the park-level monthly carbon emission monitoring results includes: The gradient boosting decision tree algorithm is used to process the model dataset and calculate the importance score of each input feature in the process of predicting monthly carbon emissions. Based on a preset importance score threshold, features with an importance score higher than the importance score threshold are selected from all input features and used as key features. Using the selected key features as input variables and historical monthly carbon emission data as target output variables, the gradient boosting decision tree algorithm is used to train the model and construct a park-level monthly carbon emission monitoring model. Using the trained monitoring model, input new monthly business activity data, and output the corresponding monthly carbon emission calculation results at the park level.
[0013] As a preferred embodiment of the park-level monthly carbon emission monitoring method described in this invention, the step of training the model using the gradient boosting decision tree algorithm includes: Set the learning rate, the maximum depth of the tree, and the subsampling rate; Based on the aforementioned key features and historical monthly carbon emission data, multiple decision trees are iteratively generated, and each decision tree is fitted with the prediction residual of the preceding decision tree. The model parameters are updated according to the gradient direction of the loss function until the prediction error of the model on the validation set converges.
[0014] As a preferred embodiment of the monthly carbon emission monitoring method for industrial parks described in this invention, the step of obtaining the annual carbon emissions of the industrial park includes: Collect annual consumption data for electricity, natural gas, gasoline, and diesel within the park; Multiply the annual consumption data of each type of energy by its corresponding carbon emission factor; The total annual carbon emissions of the park are obtained by summing the carbon emission calculations for all energy types.
[0015] Secondly, the present invention provides a park-level monthly carbon emission monitoring system, comprising: The annual data acquisition module is used to acquire data on the park's annual carbon emissions and candidate influencing factors. The key factor screening module is used to calculate and compare the correlation between annual carbon emissions and candidate influencing factor data using the Pearson correlation analysis algorithm to obtain key factors that are highly correlated with carbon emissions. The monthly economic data generation module is used to generate a simulated monthly economic indicator data sequence by using a conditional generative adversarial network, taking the park's annual economic indicator data as conditional constraint information, and combining it with random noise. The annual carbon emission decomposition module is used to evaluate the feature importance of various indicators related to the annual carbon emission amount using the random forest algorithm. Based on the importance score obtained from the evaluation, the annual carbon emission amount is weighted and decomposed into monthly weights to obtain the historical monthly carbon emission data. The multi-source data fusion module is used to integrate monthly economic indicator data sequences, historical monthly carbon emission data, and key factors highly correlated with carbon emissions to form a model dataset. The monthly carbon emission prediction module is used to extract features from the model dataset using the gradient boosting decision tree algorithm, and uses the extracted important feature data as model input to construct a park-level monthly carbon emission monitoring model and output the park-level monthly carbon emission monitoring results.
[0016] Thirdly, the present invention provides an electronic device, comprising: Memory, used to store programs; A processor is configured to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the park-level monthly carbon emission monitoring method.
[0017] Fourthly, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the park-level monthly carbon emission monitoring method.
[0018] The beneficial effects of this invention are as follows: By using the Pearson correlation analysis algorithm to calculate and compare the correlation between annual carbon emissions and candidate influencing factors, highly correlated key factors are selected, achieving accurate identification of the core driving factors affecting carbon emissions in the industrial park and providing scientific feature input for subsequent modeling. By using a conditional generative adversarial network with annual economic indicator data as conditional constraints, combined with random noise to generate simulated monthly economic indicator data sequences, an effective expansion from low-frequency annual data to high-frequency monthly data is achieved, solving the bottleneck problem of missing economic data in monthly carbon emission monitoring. By using the random forest algorithm to evaluate feature importance, annual carbon emissions are weighted and split according to the importance scores of each business activity indicator, achieving reasonable reconstruction of historical monthly carbon emissions and providing high-quality time-series label data for model training. By using the gradient boosting decision tree algorithm to extract features and model the model dataset constructed from multi-source data, high-precision prediction of monthly carbon emissions at the industrial park level is achieved, constructing a complete, closed-loop, and dynamically adaptable intelligent carbon emission monitoring system. Attached Figure Description
[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the basic process of a park-level monthly carbon emission monitoring method provided in one embodiment of the present invention.
[0020] Figure 2This is a comparison chart of monthly carbon emission monitoring results for a park-level monthly carbon emission monitoring method provided in one embodiment of the present invention. Detailed Implementation
[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0022] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for monitoring monthly carbon emissions at the park level is provided, comprising: S100: Obtain data on the park's annual carbon emissions and candidate influencing factors; S200: Using the Pearson correlation analysis algorithm, the correlation between annual carbon emissions and candidate influencing factor data is calculated and compared to obtain key factors that are highly correlated with carbon emissions. S300: By using a conditional generative adversarial network, the annual economic indicator data of the park is used as the conditional constraint information, and combined with random noise, a simulated monthly economic indicator data sequence is generated. S400: The random forest algorithm is used to assess the feature importance of various indicators related to annual carbon emissions. Based on the importance scores obtained from the assessment, the annual carbon emissions are weighted and split into monthly data to obtain the historical monthly carbon emission data. S500: It integrates monthly economic indicator data series, historical monthly carbon emission data, and key factors highly correlated with carbon emissions to form a model dataset; S600: Use the gradient boosting decision tree algorithm to extract features from the model dataset, use the extracted important feature data as model input, construct a park-level monthly carbon emission monitoring model, and output the park-level monthly carbon emission monitoring results.
[0023] It should be noted that existing carbon emission monitoring methods face a series of challenges during operation. These challenges mainly manifest in the fact that key economic and energy activity data are mostly collected and released on an annual or quarterly basis, making it difficult to obtain continuous, high-frequency monthly data, thus hindering the realization of refined and dynamic monitoring of carbon emissions. Manual reporting or estimation methods are prone to introducing errors, making it difficult to guarantee data integrity and accuracy. Traditional statistical models have limited ability to characterize nonlinear relationships and complex influencing factors, making it difficult to adapt to the dynamic changes in carbon emissions at the park level. At the same time, due to the lack of effective label data, machine learning models face difficulties in training and validating on a monthly scale, resulting in insufficient prediction accuracy and failing to meet the timeliness and reliability requirements of actual management.
[0024] Therefore, addressing the issues of untimely, inaccurate, and inconvenient monitoring of carbon emissions at the park level in existing carbon emission monitoring methods, this paper constructs a complete intelligent monthly carbon emission monitoring method at the park level through steps S100-S600. This method integrates techniques such as Pearson correlation analysis, conditional generative adversarial networks (CGAN), random forest feature importance assessment, and gradient boosting decision trees (XGBoost). It effectively overcomes the problems of monitoring lag and low accuracy caused by the lack of high-frequency activity data, blank historical monthly emission data, and insufficient prediction capabilities of traditional models in existing technologies. This method not only realizes the dimensionality increase and information reconstruction of data frequency from annual to monthly, improving data integrity and modeling quality, but also significantly improves the accuracy and intelligence level of monthly carbon emission prediction through multi-source data fusion and machine learning modeling, providing reliable technical support for parks to achieve refined and dynamic carbon management.
[0025] Example 2, this is an embodiment of the present invention, which provides a method for monitoring monthly carbon emissions at the park level based on the previous embodiment, including: In this embodiment of the application, obtaining the annual carbon emissions of the park in step S100 includes: Collect annual consumption data for electricity, natural gas, gasoline, and diesel within the park; Multiply the annual consumption data of each type of energy by its corresponding carbon emission factor; The total annual carbon emissions of the park are obtained by summing the carbon emission calculations for all energy types.
[0026] In this embodiment of the application, the carbon emission factor is determined based on the standard carbon emission coefficient published by the state or industry to ensure the standardization and comparability of the accounting results.
[0027] In this embodiment of the application, the key factor screening step S200 includes using the Pearson correlation analysis algorithm to calculate the linear correlation coefficient between annual carbon emissions and candidate influencing factors, and screening out highly correlated key factors by setting a threshold for the absolute value of correlation.
[0028] In an optional implementation, the key factor screening step in step S200 can also calculate the mutual information value between each candidate influencing factor and the annual carbon emissions, and select variables with a correlation higher than the preset mutual information threshold as key factors for subsequent model construction.
[0029] In an optional implementation, the key factor screening step in step S200 can also construct a LASSO regression model with annual carbon emissions as the dependent variable and all candidate influencing factors as independent variables. The regression coefficients of irrelevant variables are automatically compressed to zero through regularization coefficients, and the influencing factors with non-zero coefficients are identified as key factors for subsequent model construction.
[0030] In this embodiment of the application, the key factors highly correlated with carbon emissions obtained in step S200 include: The annual carbon emissions are combined with the annual data of each candidate influencing factor to form a data set; For each data combination, calculate the corresponding Pearson correlation coefficient; Pre-set a threshold for determining the absolute value of correlation; The absolute value of the Pearson correlation coefficient corresponding to each data combination is compared with the judgment threshold. Candidate influencing factors whose absolute value of the Pearson correlation coefficient is not lower than the determination threshold are identified as key factors.
[0031] In this application embodiment, the candidate influencing factors include the park's total annual energy consumption, annual electricity consumption, annual electricity sales, annual average temperature, and annual population size.
[0032] In this embodiment of the application, the Pearson correlation coefficient Calculated using the following formula: in, and Representing annual carbon emissions and a candidate influencing factor in the [number]th [year], respectively. The values for each year, and These are their means, This represents the number of years. This coefficient is used to quantify the degree of linear correlation between two variables.
[0033] In this embodiment of the application, the monthly economic indicator data generation step S300 includes using a conditional generative adversarial network (CGAN), taking annual economic indicators as conditional inputs, combining random noise, and generating a simulated monthly economic indicator sequence that conforms to the real data distribution through adversarial training between the generator and the discriminator.
[0034] In an optional implementation, the monthly economic indicator data generation step in step S300 can also construct a conditional variational autoencoder model, using the park's annual economic indicator data as conditional input, learning the potential distribution of real monthly economic data through the encoder, and then the decoder samples from the distribution to generate a simulated monthly economic indicator data sequence that conforms to the characteristics of time series.
[0035] In an optional implementation, the monthly economic indicator data generation step in step S300 can further decompose the historical annual economic indicator data into a time series, generate monthly trend components through interpolation, and superimpose the extracted seasonal patterns and random disturbances that conform to statistical characteristics to synthesize a simulated monthly economic indicator data series with reasonable time sequence regularity.
[0036] In this embodiment of the application, step S300, generating a simulated monthly economic indicator data sequence, includes: Using the park's annual economic indicators as input, a conditional generative adversarial network containing a generator and a discriminator is constructed. In each iteration, the generator parameters are fixed, and the discriminator is trained by inputting real monthly economic indicator data and conditional input. In each iteration, the discriminator parameters are fixed, and random noise and conditional labels are input to train the generator; The training process for the discriminator and generator is performed alternately to continuously optimize the parameters of the discriminator and generator; Using the trained generator, combined with new annual economic indicator data as conditional input and random noise, simulated monthly economic indicator data sequences are generated.
[0037] In this embodiment, the training process of the conditional generative adversarial network aims to minimize the difference between the generated data and the real data distribution, ensuring that the generated monthly economic indicator data has real data characteristics and time series patterns.
[0038] In this embodiment of the application, the training objective of the conditional generative adversarial network is achieved through the following objective function: in, Indicates the discriminator, Represents a generator. This represents real monthly economic indicator data. This indicates the input conditions for annual economic indicators. Indicates random noise. Represents the distribution of real data. This represents the prior distribution of the noise. The objective function guides the adversarial training process between the generator and the discriminator, enabling the generator to produce simulated monthly data that is as consistent as possible with the real data distribution.
[0039] In this embodiment of the application, the step S400 of breaking down the annual carbon emissions into monthly amounts includes evaluating the importance score of each business activity indicator through a random forest algorithm, using the score as a weight, and then weighting and summing the monthly data of each indicator to achieve the monthly breakdown of the annual carbon emissions.
[0040] In an optional implementation, the annual carbon emissions breakdown into monthly steps in step S400 can also involve principal component analysis of the annual business activity indicator dataset, extracting the first principal component, using the loading values of each indicator on the principal component as its weights, multiplying them by the monthly data of the corresponding indicators, and then weighted summing to obtain the carbon emissions data for each historical month.
[0041] In an optional implementation, in step S400, the annual carbon emissions are broken down into monthly data. In this step, domain experts can be organized to compare the impact of each business activity indicator pairwise and construct a judgment matrix. The subjective weight of each indicator is determined by calculating the feature vector. The weight is then multiplied by the monthly data of the corresponding indicator and weighted and summed to obtain the carbon emissions data for each historical month.
[0042] In this embodiment of the application, step S400 obtains historical carbon emission data for each month, including: Obtain annual carbon emission data at the park level, as well as related annual business activity indicator datasets; The annual business activity indicator dataset was processed using the random forest algorithm to calculate the importance score of each business activity indicator in the process of predicting annual carbon emissions. The importance score of each business activity indicator will be used as the weight of the current business activity indicator. The monthly data for each business activity indicator is multiplied by its corresponding importance score to obtain the weight. By summing the weighted monthly data of all business activity metrics, we can obtain the historical carbon emission data for each month.
[0043] In this application embodiment, business activity indicators include monthly electricity consumption, monthly water consumption, monthly logistics and transportation volume, monthly production man-hours, and monthly equipment operating time.
[0044] In this embodiment of the application, the monthly carbon emission prediction model construction step S600 includes using the Gradient Boosting Decision Tree (XGBoost) algorithm to evaluate and screen the feature importance of the fused model dataset, and train a park-level monthly carbon emission monitoring model.
[0045] In an optional implementation, the monthly carbon emission prediction model construction step S600 can also use multi-dimensional time series data composed of integrated monthly economic indicators, historical carbon emissions, and key factors as input to construct an LSTM neural network model. By training and learning the time-series dependence of carbon emissions, the model can output the predicted carbon emissions for each future month at the park level.
[0046] In an optional implementation, the monthly carbon emission prediction model construction step S600 can also use the selected key features as input variables and historical monthly carbon emission data as target output to construct a support vector regression model. By selecting an appropriate kernel function to fit the nonlinear relationship, a park-level monthly carbon emission monitoring model can be trained and used to predict the carbon emission of future months.
[0047] In this embodiment of the application, step S600 outputs the monthly carbon emission monitoring results at the park level, including: The gradient boosting decision tree algorithm is used to process the model dataset and calculate the importance score of each input feature in the process of predicting monthly carbon emissions. Based on a preset importance score threshold, features with an importance score higher than the importance score threshold are selected from all input features and used as key features. Using the selected key features as input variables and historical monthly carbon emission data as target output variables, the gradient boosting decision tree algorithm is used to train the model and construct a park-level monthly carbon emission monitoring model. Using the trained monitoring model, input new monthly business activity data, and output the corresponding monthly carbon emission calculation results at the park level.
[0048] In this embodiment of the application, step S600, which involves training the model using the gradient boosting decision tree algorithm, includes: Set the learning rate, the maximum depth of the tree, and the subsampling rate; Based on the aforementioned key features and historical monthly carbon emission data, multiple decision trees are iteratively generated, and each decision tree is fitted with the prediction residual of the preceding decision tree. The model parameters are updated according to the gradient direction of the loss function until the prediction error of the model on the validation set converges.
[0049] In this embodiment, the gradient boosting decision tree algorithm is implemented using XGBoost, whose objective function includes a loss term and a regularization term to prevent model overfitting and improve generalization ability.
[0050] In this embodiment, the gradient boosting decision tree algorithm is implemented using XGBoost, and the goal of its model training is to minimize the following objective function: in, Indicates the first The actual monthly carbon emissions of each sample This represents the model's predicted value. For loss functions (such as mean squared error RMSE). Indicates the first A decision tree, This is the complexity regularization term for the tree, used to prevent overfitting. The objective function guides the model to reduce prediction error while maintaining structural simplicity, thus achieving good generalization ability.
[0051] In this embodiment of the application, after outputting the monthly carbon emission monitoring results at the park level, the method further includes: The newly acquired monthly business activity data of the park and its corresponding carbon emission calculation results are added as new samples to the model dataset; The park-level monthly carbon emission monitoring model is periodically trained using the updated model dataset to achieve dynamic updates and continuous optimization of the model.
[0052] This mechanism enables the model to adapt to changes in park operation patterns, energy structure, or external environment, maintaining the accuracy of long-term forecasts.
[0053] Example 3, referring to Figure 2 This is one embodiment of the present invention, which provides a method for monitoring monthly carbon emissions at the park level. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through specific implementation methods and implementation effects.
[0054] The specific details of this embodiment are as follows: 1. Experimental Environment This embodiment takes a zero-carbon industrial park in Yunnan Province as the research object, selecting its historical data from 2019 to 2023 as the modeling basis. The experimental environment includes: Hardware configuration: Intel Xeon E5-2680 v4 CPU @ 2.40GHz, 64GB RAM, NVIDIA Tesla T4 GPU, running environment Ubuntu 20.04 operating system.
[0055] Software platform: Python 3.8 programming language, with major dependencies including NumPy, Pandas, Scikit-learn, XGBoost, TensorFlow 2.8, and Statsmodels.
[0056] Data source: The annual carbon emissions (1200~1500 tCO2) are calculated based on the annual consumption of electricity, natural gas, gasoline and diesel in the park and the corresponding carbon emission factors. Annual GDP (800-1 billion yuan) and annual electricity consumption (5-6 million kWh) are from the park's annual economic report and energy management system; High-frequency data such as monthly electricity consumption (300,000 to 700,000 kWh), electricity sales (280,000 to 650,000 kWh), and average temperature (10 to 28°C) are obtained from smart meters, marketing systems, and weather stations.
[0057] Model parameters: Both the CGAN generator and discriminator are 3-layer fully connected networks with ReLU and Leaky ReLU activation functions, respectively, and the training iterations are 1000 rounds. The XGBoost model is configured with a maximum tree depth of 6, a learning rate of 0.1, a subsampling rate of 0.8, and a feature importance threshold of 0.05.
[0058] 2. Experimental Procedure The experiment was conducted according to the following steps: Step 1: Annual Data Acquisition and Key Factor Screening The annual consumption of electricity, natural gas, gasoline, and diesel in the industrial park from 2019 to 2023 was collected and weighted by combining the national carbon emission factor to calculate the total annual carbon emissions. Subsequently, Pearson correlation analysis was used to calculate the correlation coefficient between carbon emissions and candidate factors such as GDP, electricity consumption, electricity sales, and temperature. A threshold of 0.8 for the absolute value of the correlation was set, and GDP and electricity consumption were selected as key influencing factors.
[0059] Step Two: Monthly Economic Data Generation and Carbon Emission Decomposition A conditional generative adversarial network (CGAN) is constructed, using annual GDP as input, to train a generator that simulates monthly GDP sequences. Simultaneously, a random forest algorithm is used to evaluate the importance scores of business indicators such as electricity consumption and sales to annual carbon emissions, and these scores are used as weights to weight and sum the monthly data for each indicator, thus achieving a monthly breakdown of annual carbon emissions and obtaining historical carbon emission data for each month from 2019 to 2023.
[0060] Step 3: Model Building and Prediction Output The generated monthly GDP, broken down monthly carbon emissions, and key factors (electricity consumption, temperature, etc.) are time-aligned and fused to construct a model dataset. The XGBoost algorithm is used to perform feature importance analysis on the dataset, selecting features with an importance score higher than 0.05 as input. Historical monthly carbon emissions are used as output to train the monthly carbon emission monitoring model. Time-series cross-validation is employed during model training to prevent overfitting. After training, business data for each month of 2024 is input, and the corresponding monthly carbon emission prediction results are output.
[0061] 3. Experimental Results The modeling evaluation results of this method on historical data from 2019 to 2023 show that the coefficient of determination (R²) reaches 0.963 and the mean absolute percentage error (MAPE) is 5.2%, indicating that the model has high goodness of fit and prediction accuracy. Figure 2 As shown, the monitoring results can effectively capture the peak carbon emissions caused by the increase in air conditioning load in summer, and have a more dynamic response capability compared with the traditional average allocation method. The monthly GDP generated by CGAN and the carbon emission sequences decomposed by random forest successfully compensate for the lack of high-frequency data, providing a reliable technical tool for the park to achieve refined carbon management. The model has been iterated and updated quarterly since its launch, maintaining predictive stability, and has been applied to the park's carbon quota allocation and emission reduction strategy optimization.
[0062] Example 4 is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a park-level monthly carbon emission monitoring system.
[0063] It should be noted that the technical solution of the park-level monthly carbon emission monitoring system is based on the same concept as the technical solution of the park-level monthly carbon emission monitoring method described above. For details not described in detail in the technical solution of the park-level monthly carbon emission monitoring system in this embodiment, please refer to the description of the technical solution of the park-level monthly carbon emission monitoring method described above.
[0064] This embodiment describes a park-level monthly carbon emission monitoring system, including: The annual data acquisition module is used to acquire data on the park's annual carbon emissions and candidate influencing factors. The key factor screening module is used to calculate and compare the correlation between annual carbon emissions and candidate influencing factor data using the Pearson correlation analysis algorithm to obtain key factors that are highly correlated with carbon emissions. The monthly economic data generation module is used to generate a simulated monthly economic indicator data sequence by using a conditional generative adversarial network, taking the park's annual economic indicator data as conditional constraint information, and combining it with random noise. The annual carbon emission decomposition module is used to evaluate the feature importance of various indicators related to the annual carbon emission amount using the random forest algorithm. Based on the importance score obtained from the evaluation, the annual carbon emission amount is weighted and decomposed into monthly weights to obtain the historical monthly carbon emission data. The multi-source data fusion module is used to integrate monthly economic indicator data sequences, historical monthly carbon emission data, and key factors highly correlated with carbon emissions to form a model dataset. The monthly carbon emission prediction module is used to extract features from the model dataset using the gradient boosting decision tree algorithm, and uses the extracted important feature data as model input to construct a park-level monthly carbon emission monitoring model and output the park-level monthly carbon emission monitoring results.
[0065] This embodiment also provides an electronic device applicable to a park-level monthly carbon emission monitoring method, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a campus-level monthly carbon emission monitoring method as described in the above embodiments.
[0066] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for monitoring monthly carbon emissions at the park level as proposed in the above embodiments.
[0067] The storage medium proposed in this embodiment and the method for implementing a monthly carbon emission monitoring at the park level proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0068] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A park-level monthly carbon emission monitoring method, characterized in that, The method comprises the following steps: obtaining annual carbon emissions and candidate influencing factor data of the park; performing correlation calculation and comparison on the annual carbon emissions and the candidate influencing factor data by a Pearson correlation analysis algorithm to obtain key factors highly correlated with the carbon emissions; generating simulated monthly economic indicator data sequences by a conditional generative adversarial network, taking annual economic indicator data of the park as conditional constraint information and combining random noise; performing feature importance evaluation on each index related to the annual carbon emissions by a random forest algorithm, taking the importance scores obtained by the evaluation as weights, and weighting and splitting the annual carbon emissions monthly to obtain historical monthly carbon emissions data; integrating the monthly economic indicator data sequences, the historical monthly carbon emissions data and the key factors highly correlated with the carbon emissions to form a model data set; performing feature extraction on the model data set by a gradient boosting decision tree algorithm, taking the extracted important feature data as model input, constructing a park-level monthly carbon emission monitoring model, and outputting park-level monthly carbon emission monitoring results.
2. The park-level monthly carbon emission monitoring method of claim 1, wherein: The key factors highly correlated with the carbon emissions are obtained by: combining the annual carbon emissions with the annual data of each candidate influencing factor to form data combinations; calculating the Pearson correlation coefficients corresponding to each data combination; pre-setting a correlation absolute value judgment threshold; comparing the absolute values of the Pearson correlation coefficients corresponding to each data combination with the judgment threshold; determining the candidate influencing factors with the absolute values of the Pearson correlation coefficients not lower than the judgment threshold as the key factors.
3. The park-level monthly carbon emission monitoring method of claim 1 or 2, wherein: The simulated monthly economic indicator data sequences are generated by: taking the annual economic indicator data of the park as conditional input to construct a conditional generative adversarial network comprising a generator and a discriminator; in each iteration, fixing the generator parameters, inputting real monthly economic indicator data and conditional input to train the discriminator; in each iteration, fixing the discriminator parameters, inputting random noise and conditional labels to train the generator; alternately performing the training processes of the discriminator and the generator to continuously optimize the parameters of the discriminator and the generator; using the trained generator, combining new annual economic indicator data as conditional input and random noise to generate simulated monthly economic indicator data sequences.
4. The park-level monthly carbon emission monitoring method of claim 3, wherein: The historical monthly carbon emissions data are obtained by: obtaining park-level annual carbon emissions data and an annual business activity indicator data set related thereto; processing the annual business activity indicator data set by a random forest algorithm to calculate the importance scores of each business activity indicator in the process of predicting annual carbon emissions; taking the importance scores of each business activity indicator as the weights of the current business activity indicator; multiplying the monthly data of each business activity indicator by the corresponding importance score as the weight to obtain the result; adding all the weighted monthly data of the business activity indicators to obtain the historical monthly carbon emissions data.
5. The park-level monthly carbon emission monitoring method of claim 4, wherein: The park-level monthly carbon emission monitoring results are output by: processing the model data set by a gradient boosting decision tree algorithm to calculate the importance scores of each input feature in the process of predicting monthly carbon emissions; Based on a preset importance score threshold, features with an importance score higher than the importance score threshold are selected from all input features and used as key features. Using the selected key features as input variables and historical monthly carbon emission data as target output variables, the gradient boosting decision tree algorithm is used to train the model and construct a park-level monthly carbon emission monitoring model. Using the trained monitoring model, input new monthly business activity data, and output the corresponding monthly carbon emission calculation results at the park level.
6. The park-level monthly carbon emission monitoring method of claim 5, wherein: The steps for training the model using the gradient boosting decision tree algorithm include: Set the learning rate, the maximum depth of the tree, and the subsampling rate; Based on the aforementioned key features and historical monthly carbon emission data, multiple decision trees are iteratively generated, and each decision tree is fitted with the prediction residual of the preceding decision tree. The model parameters are updated according to the gradient direction of the loss function until the prediction error of the model on the validation set converges.
7. The park-level monthly carbon emission monitoring method of claim 6, wherein: The acquisition of the park's annual carbon emissions includes: Collect annual consumption data for electricity, natural gas, gasoline, and diesel within the park; Multiply the annual consumption data of each type of energy by the corresponding carbon emission factor; The total annual carbon emissions of the park are obtained by summing the carbon emission calculations for all energy types.
8. A park-level monthly carbon emission monitoring system applying the method of any one of claims 1-7, characterized in that, include: The annual data acquisition module is used to acquire data on the park's annual carbon emissions and candidate influencing factors. The key factor screening module is used to calculate and compare the correlation between annual carbon emissions and candidate influencing factor data using the Pearson correlation analysis algorithm to obtain key factors that are highly correlated with carbon emissions. The monthly economic data generation module is used to generate a simulated monthly economic indicator data sequence by using a conditional generative adversarial network, taking the park's annual economic indicator data as conditional constraint information, and combining it with random noise. The annual carbon emission decomposition module is used to evaluate the feature importance of various indicators related to the annual carbon emission amount using the random forest algorithm. Based on the importance score obtained from the evaluation, the annual carbon emission amount is weighted and decomposed into monthly weights to obtain the historical monthly carbon emission data. The multi-source data fusion module is used to integrate monthly economic indicator data sequences, historical monthly carbon emission data, and key factors highly correlated with carbon emissions to form a model dataset. The monthly carbon emission prediction module is used to extract features from the model dataset using the gradient boosting decision tree algorithm, and uses the extracted important feature data as model input to construct a park-level monthly carbon emission monitoring model and output the park-level monthly carbon emission monitoring results.
9. An electronic device, comprising: include: Memory, used to store programs; A processor for loading the program to perform the steps of the method as claimed in any one of claims 1-7.
10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.