Present situation of China’s carbon emissions based on IEA CO2 Emissions Database

Since the industrial revolution, the power industry has experienced the development process from steam power generation to thermal power generation, and then to the diversified energy structure. The energy structure of the world’s power industry is currently undergoing a major transition, and renewable energy sources including hydropower, wind energy, and solar energy are progressively contributing to the production of electricity. China’s power sector has also accomplished a great deal in recent years. Not only does it have some of the highest installed power capacity and power generation in the world, but it has also made significant strides toward the development of clean energy. In terms of the structure of power production, thermal power generation dominates China’s power sector. Nevertheless, the share of clean energy power generation is growing annually. Gas-fired power generation and coal-fired power generation make up the bulk of thermal power generation, with coal-fired power generation holding a leading position. Nonetheless, the share of renewable energy, such as solar and wind power, in power generation has grown annually due to the ongoing advancements in clean energy technology and cost reductions. Nuclear power is a clean, effective energy source that contributes significantly to China’s electricity output.

With China’s economy expanding quickly and the country’s citizens enjoying higher living standards, there is an increasing need for power consumption. The primary sectors of electricity use are industrial, commercial, and residential, with industrial electricity making up a sizable share of all three. The rapidly evolving fields of electric vehicles and smart houses are causing a progressive shift in the structure of power consumption as well [24–26]. This study makes use of the IEA CO₂ Emissions Database to more precisely analyze the current state of carbon emissions in China’s power industry. The database offers comprehensive information on carbon dioxide emissions from both domestic and international energy operations, which is very helpful in assessing the carbon emissions of the power sector. China’s power industry’s carbon emissions from 2016 to 2022 are displayed in Fig 1. It demonstrates that China’s power industry’s carbon emissions have been trending toward first rising and then falling in recent years. Between 2016 and 2019, the power industry’s carbon emissions increased, but at a slower rate due to changes in the energy structure and the rise in power demand. 2020 saw a decrease in the demand for power due to the global COVID-19 pandemic and changes to the domestic economic structure, which led to a negative increase in carbon emissions. In 2021, with the economic recovery, carbon emissions rebounded slightly, but the growth rate slowed down significantly. By 2022, carbon emissions will decline again, which shows that China’s power industry has achieved positive results in emission reduction.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Overview of carbon emissions of power industry in China from 2016 to 2022.

https://doi.org/10.1371/journal.pone.0316537.g001

In China’s power industry, thermal power generation—particularly coal-fired power generation—is the primary source of carbon emissions. As a result, energy structure optimization is crucial for lowering carbon emissions in the power sector. China has successfully decreased the power industry’s carbon emission intensity in recent years by accelerating the development of clean energy and encouraging the conversion of coal-fired power to ultra-low emission sources. Regional variations in the power industry’s carbon emissions are evident. The eastern region’s strong demand for electricity and established economy have resulted in comparatively high carbon emissions. According to the research of Xie, Wang [27], due to the developed economy and dense population, the eastern coastal areas had relatively high-power demand and relatively large carbon emissions. The power consumption of eastern provinces such as Jiangsu and Guangdong have been in the forefront of the country for a long time, which was closely related to its economic development level and industrialization level. In the research of Zhu, Wang [28], it was considered that the proportion of thermoelectric production in the eastern region increases due to the increase of heating demand in winter, which led to the corresponding increase of carbon emission level. Comparatively speaking, Zheng, Deng [29] considered that the western region’s carbon emission level was kept at a low level due to the relatively backward economic development, low total electricity demand and abundant clean energy resources. In western provinces such as Qinghai and Yunnan, the seasonal fluctuation of power demand had little impact on the overall carbon emissions. In these areas, due to the abundant solar and wind energy resources in summer, they can better cope with the increase in power demand while maintaining low-carbon emissions. Therefore, the power structure and carbon emission characteristics in the western region were in sharp contrast with those in the eastern region.

Carbon emission cost accounting of electric power enterprises

The power industry is progressively shifting away from the conventional method of coal-fired power generation and toward a more diverse and clean direction due to the severity of climate change on a worldwide scale and China’s strategic aim of “peak carbon dioxide emissions and carbon neutrality”. The accounting of carbon emission costs has developed into a crucial metric in this process for assessing the degree of green development of power companies. In terms of installed capacity and power generation, China’s power industry is among the best in the world. Nonetheless, a significant amount of power is still generated by thermal sources, particularly coal-fired power. The power industry’s energy structure is gradually being optimized due to the rapid development of renewable energy sources like solar and wind power [30, 31].

The accounting content of carbon emission cost of power enterprises covers many aspects, the first is the accurate calculation of carbon emission. This requires enterprises to use scientific calculation methods to estimate carbon emissions based on data such as power generation and coal consumption. Other greenhouse gas emissions should also be considered to fully reflect the carbon emissions in the process of power production. In addition to carbon emissions, resource loss cost is also an important part of carbon emission cost accounting. This includes the cost of procurement, transportation and storage of fossil energy such as coal. In the accounting process, it is necessary to combine resource loss with carbon emissions to reflect the resource consumption cost caused by carbon emissions. The value of environmental damage is also a part that cannot be ignored. Although this part of the cost is difficult to measure directly, it can be estimated by introducing an environmental damage assessment model. It represents the value that electric power enterprises should bear for the potential damage to the environment caused by greenhouse gas emissions.

To maintain the accuracy and scientific nature of accounting, a set of guidelines should be adhered to when verifying the carbon emission cost of power enterprises. The definability principle requires that the carbon emission cost must be clearly defined and accurately understood and applied by enterprises. The principle of measurability emphasizes that the cost of carbon emissions should be accurately measured in monetary units and reflected in the accounting statements of enterprises. In order for the carbon emission cost to accurately reflect the carbon emissions during the power production process, it must be closely tied to the production and operation activities of businesses, according to the principle of correlation. Fig 2 illustrates the specific confirmation process of carbon emission cost.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Specific confirmation process of carbon emission cost.

https://doi.org/10.1371/journal.pone.0316537.g002

In the specific accounting process, firstly, it is necessary to collect and sort out relevant data such as power generation, coal consumption and investment in environmental protection facilities of electric power enterprises. Then, carbon emissions are calculated according to these data and scientific calculation methods [32–34]. Then, carbon emissions, resource loss cost, environmental damage value, carbon emission reduction cost and carbon transaction cost are included in the accounting scope for comprehensive calculation. Finally, the accounting results are reflected in the accounting statements of enterprises, which provides strong support for the decision-making and green development of enterprises.

Data mining of enterprise carbon emissions based on machine learning

The process of using algorithms to find hidden information in huge datasets is known as data mining. This procedure, which is typically associated with computer science, is carried out using a variety of techniques, including pattern recognition, machine learning, statistics, online analytical processing, information retrieval, and expert systems. Pretreatment, integration, selection, and preparation of data are all steps in the data mining process. Its goal is to find knowledge that users are interested in, which is acceptable, understandable and applicable, and supports specific applications to find problems. There are many algorithms in data mining, including classification, regression analysis, clustering, association rules and so on.

This study uses machine learning algorithm to deeply mine the carbon emission data of power enterprises, and choose decision tree algorithm as the main tool. This algorithm is superior in dealing with regression and classification problems, and has outstanding ability in dealing with data characteristics and prediction results. Microsoft decision tree algorithm is suitable for complex classification and regression problems, and its obvious advantages are that the model is easy to explain, the results are intuitive, and it can effectively deal with nonlinear data relationships, so this algorithm can meet the complexity requirements of carbon emission data. In the data preprocessing stage, this study first collects the original data from enterprise database and official website from 2020 to 2022, including power generation, fuel consumption and investment in environmental protection facilities. Then, it cleans the data, removes the missing values and abnormal values, and fills the missing data and correct the abnormal values through mean interpolation and interpolation method. On this basis, it also unifies the data format to ensure the consistency of data from different sources to facilitate the subsequent analysis, and standardized the data, and normalized all numerical data to values in the interval of [0, 1], thus eliminating the influence brought by different dimensions. In the feature extraction process, the importance of features is evaluated by decision tree algorithm, and it selects the features that have a significant impact on carbon emissions, such as total electricity production, coal consumption and environmental protection investment. At the same time, according to the business requirements and model requirements, it has constructed some new features, such as carbon emission intensity and clean energy ratio, which are helpful to improve the prediction performance of the model. Subsequently, in the model training step, it uses the decision tree algorithm to build the model. This algorithm creates a decision tree by dividing the data set into multiple subsets, in which each node represents the test conditions of feature attributes and the branches represent the results of feature attributes. In order to prevent the model from over-fitting, it sets parameters such as the maximum depth of the tree and the minimum number of sample splits, and uses cross-validation method to evaluate the generalization ability of the model to optimize the model parameters. Finally, the model is verified by training set and test set to ensure its accuracy and stability.

In this study, decision tree algorithm, Transformer model and entropy weight -TOPSIS method are selected, which are based on their unique advantages in processing carbon emission data and comprehensive evaluation. Decision tree algorithm is a kind of supervised learning technology, which has clear decision rules and is easy to explain. It can effectively handle large-scale datasets and provide an intuitive understanding of the importance of variables. Microsoft decision tree algorithm is chosen because it is excellent in dealing with the prediction modeling of discrete and continuous attributes, can accurately divide the data feature space, and is helpful to identify the key factors affecting carbon emissions. Transformer model is an important progress in the field of natural language processing in recent years, which can deal with complex sequence data. Because of the need to deal with a large number of time-series carbon emission data, Transformer model can capture long-distance dependence and effectively deal with unstructured information in the data. Compared with the traditional time series analysis method, the Transformer model is outstanding in dealing with large-scale data and provides more accurate analysis results. Entropy weight -TOPSIS method is a comprehensive evaluation method, which combines the advantages of entropy weight method and TOPSIS method Entropy weight method can objectively determine the weight of each evaluation index, while TOPSIS method ranks the schemes by comparing the distance between them and the ideal scheme. This method can comprehensively consider multiple evaluation indexes and provide scientific evaluation results through quantitative analysis. Compared with other evaluation methods, entropy weight -TOPSIS method has strong comprehensive analysis ability and accuracy when dealing with multi-dimensional and multi-objective evaluation problems. Table 1 shows the comparative analysis:

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Method comparative analysis.

https://doi.org/10.1371/journal.pone.0316537.t001

The decision tree method is a supervised learning technique used in data mining for problems with regression and classification. It divides the dataset into smaller groups iteratively to construct a decision tree. When building a decision tree, each node stands for the feature attribute’s test condition, and each branch represents the feature attribute’s result within a given value range. A category or a certain number is stored in each leaf node. The Microsoft SQL Server Analysis Services offers the Microsoft decision tree method, a classification and regression technique that is particularly well-suited for predictive modeling of discrete and continuous attributes. By constructing a decision tree model, the algorithm divides the feature space of data instances into several disjoint subsets, and each subset corresponds to a category or regression value.

The Microsoft decision tree algorithm meticulously plans a number of breaks in the tree structure to create an effective data mining model. The method adds a new node to the model whenever it finds a substantial correlation between the input data’s columns and the predictable columns. “Feature selection” is a technique used by the Microsoft decision tree algorithm to help identify the most beneficial qualities. By preventing unnecessary characteristics from taking up processor time, all of these strategies contribute to increased performance and higher-quality analysis. The decision tree creation procedure is depicted in Fig 3. In order to assist users in extracting valuable knowledge and information from massive datasets, this technique is mostly utilized in the domains of data mining and machine learning.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Decision tree construction process in Microsoft decision tree algorithm.

https://doi.org/10.1371/journal.pone.0316537.g003

Prior to engaging in carbon emission data mining, businesses’ carbon emission-related data must be gathered and sorted. Power generation, fuel consumption, environmental protection facility operation data, and production process parameters are a few examples of these data [35, 36]. Preprocessing the data is required to guarantee the precision and efficacy of data mining. This includes data cleansing, unifying the format, handling outliers, and other procedures. Preparing the data is a crucial first step in the data mining process. This covers tasks like feature extraction, data conversion, and cleaning. Transformer architecture can handle unstructured data, including text data. Transformer architecture, a significant advancement in natural language processing in recent years, offers a fresh viewpoint on evaluating English instruction since it can manage complicated text material and identify long-term relationships in sequences. In enterprise carbon emission data mining, a large number of carbon emission data need to be collected and processed first. These data may include energy consumption, production process, equipment status and other information. Because the Transformer model can process the sequence data, it is necessary to properly preprocess these data, such as time series transformation and normalization, so that the model can be better understood and processed.

Transformer architecture is the foundation for the construction of a multi-layer encoder-decoder model. The input text data must be transformed by the encoder into a set of feature vectors, which the decoder then uses to produce the evaluation results of the students’ learning. In order to maximize the model’s performance, the parameters and structure are modified during the model-building process based on the unique requirements. The Transformer model, which is utilized to describe the dependency between any two points in the sequence, is based on a self-attention process. The following is the equation used to calculate the attention weight: (1) Q is the query matrix. K is the key matrix. V is the value matrix, and d_k is the dimension of the key vector, which is used to scale the dot product to avoid gradient disappearance or gradient explosion.

Carbon audit evaluation of power industry based on DSR model

In the power industry, the management of carbon emission cost is a multi-dimensional and whole-process systematic project, and the selection of its indicators must take into account the comprehensive consideration before, during and after. First, considering pre-control, the choice of carbon emission cost indicators should be made with the low-carbon development strategy of the business in mind. This will guarantee that the indicators chosen can effectively steer the business in the direction of a low-carbon future. This requires enterprises to consider not only economic benefits, but also the carbon emission cost of projects when formulating investment strategies, and give priority to those emission reduction technologies or projects with significant low-carbon benefits and high cost-benefit ratio. Meanwhile, enterprises need to pay close attention to national and local laws and regulations on carbon emissions to ensure that the selected indicators meet the policy requirements and avoid compliance risks.

In the process control stage, real-time monitoring and dynamic adjustment of carbon emission cost is very important. Enterprises need to establish a perfect carbon emission monitoring system and use advanced information technology to monitor carbon emissions in real time and accurately. On this basis, enterprises should also regularly carry out cost-benefit analysis, compare the actual effect of emission reduction measures with the expected goals, and adjust emission reduction strategies in time to ensure effective control of emission reduction costs [37, 38]. In addition, enterprises need to strengthen risk management, identify and evaluate potential risks related to carbon emission costs, such as policy changes and market price fluctuations, and formulate effective countermeasures to ensure the stable operation of enterprises.

In the post-event control stage, the performance evaluation and continuous improvement of carbon emission cost is an indispensable part. Enterprises should establish a scientific performance evaluation system to quantitatively evaluate the effect of emission reduction, and take the evaluation results as the basis for improvement. Through the feedback mechanism, enterprises can timely understand the effectiveness and shortcomings of emission reduction work, and provide direction for subsequent improvement. At the same time, enterprises should actively fulfill their social responsibilities, compile and publish social responsibility reports and mass satisfaction questionnaires, disclose carbon emission cost information to the public, and enhance the transparency and social responsibility of enterprises. This will not only help to enhance the brand image of enterprises, but also set a low-carbon development model for other enterprises. Table 2 shows the summary of the three dimensions: before, during and after.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Three control dimensions: Before, during and after.

https://doi.org/10.1371/journal.pone.0316537.t002

DSR model is put forward by the United Nations Commission on Sustainable Development, which aims to understand and evaluate the impact of social and economic development on the natural environment through three key dimensions (drivers, state and response), as well as human responses and countermeasures to these impacts. In the DSR model, the drivers usually refer to the potential causes or forces that lead to environmental changes in the socio-economic system. The state refers to the state or characteristics of the system at a certain point in time, reflecting the results of the system affected by the drivers. Response usually refers to the policies, technologies and management measures taken by the government, enterprises, the public and other subjects to cope with environmental changes. The DSR model diagram is shown in Fig 4.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. DSR model diagram.

https://doi.org/10.1371/journal.pone.0316537.g004

DSR model aims to understand and evaluate the impact of social and economic development on the natural environment through three key dimensions-Drivers, State and response, as well as the human response and response measures to these impacts. In the DSR model, the driving factors refer to the potential causes or forces that cause environmental changes. In the power industry, these factors usually include the adjustment of energy structure, technological progress and policy environment. For example, the transition from fossil fuel to renewable energy and the application of carbon capture and storage technology are all important driving factors. By analyzing these factors, we can identify the main forces that affect carbon emissions and formulate targeted carbon emission reduction measures. In China, the support of national policies for the development of clean energy is regarded as a key driver for the transformation of the power industry. State represents the state or characteristics of the system at a specific point in time, reflecting the results of the system being affected by driving factors. In carbon audit evaluation, status mainly refers to the carbon emission status and energy efficiency of enterprises, such as carbon dioxide, sulfur dioxide and soot emissions, coal consumption rate of power supply and comprehensive auxiliary power consumption rate of thermal power business. These state indicators can effectively reflect the carbon emission level and energy utilization of enterprises at different time points. Response refers to the policies, technologies and management measures taken by the government, enterprises and the public to cope with environmental changes. In the power industry, these response measures include the development and application of clean energy, the implementation of carbon capture and storage technology, ultra-low emission transformation and environmental protection investment. Through the analysis of these response measures, people can evaluate the positive actions of enterprises in reducing carbon emissions and improving energy efficiency and their effects. When applying DSR model to carbon audit evaluation of power industry, firstly, identify and quantify the driving factors of power enterprises, such as economic indicators such as total assets, total profits and power sales, as well as production indicators such as power generation and coal consumption. Then, according to the analysis results of driving factors, the carbon emission status of enterprises is evaluated, and environmental performance and energy consumption are concerned to judge the current situation of enterprises in carbon emission reduction and energy efficiency improvement. Finally, it analyzes the response measures taken by enterprises in dealing with carbon emissions, and evaluates their effectiveness, including management strategies and related indicators of low-carbon governance.

Through the above steps, DSR model can systematically reveal the changing mechanism of carbon emission in power industry under the action of external driving factors, and how all parties can promote the adoption of positive measures through feedback mechanism after the carbon emission status changes, and further promote the development of the whole industry in the direction of low carbon and environmental protection through virtuous circle mechanism. Based on the actual cases of two electric power enterprises in China, this study constructs a carbon audit evaluation index system with 18 indexes, and analyzes it by using entropy weight -TOPSIS comprehensive evaluation method. This process not only shows the practical application of DSR model in carbon audit evaluation, but also provides scientific basis and optimization direction for carbon emission reduction strategy of power enterprises.

Based on the existing DSR model, time series analysis method is introduced to capture the dynamic changes of carbon emissions in power industry. Time series analysis is to model and predict the regularity of time series data, thus reflecting the trend and periodic changes of data over time. In this study, a time series regression model (such as ARIMA model) is used to analyze the carbon emission data of power enterprises from 2020 to 2022 to identify the dynamic changes and key driving factors of carbon emissions at different time points. The time series analysis results of carbon emission data of China Resources Power and Datang International Power Generation Co., Ltd. from 2020 to 2022 is shown in Table 3. According to the time series analysis results in Table 3, it shows that the comprehensive evaluation index of China Resources Power presents an increasing trend year by year, from 0.5458 in 2020 to 0.6270 in 2022. This shows that the company has achieved remarkable results in promoting low-carbon energy and reducing the proportion of thermal power. The increase of driving factor D shows that the driving force of economy and policy is gradually enhanced, the promotion of state factor S shows the continuous improvement of carbon emission control, and the steady growth of response factor R reflects the continuous investment and optimization of low-carbon governance measures. In contrast, the comprehensive evaluation index of Datang International Power Generation Co., Ltd. fluctuates greatly, especially in 2021, which is related to the company’s retrogression in sulfur dioxide and carbon dioxide emission reduction. However, the company’s index rebounded in 2022, indicating that its carbon emission management effect has improved after adjusting its management strategy and strengthening investment in environmental protection. This result further proves the importance of time series analysis in capturing the dynamic changes of carbon emission management.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Time series analysis results of carbon emission data of China Resources Power and Datang International Power Generation Co., Ltd. from 2020 to 2022.

https://doi.org/10.1371/journal.pone.0316537.t003

D-S mechanism reveals the internal logic of how the carbon emission state of power industry changes under the action of external drivers. The S-R mechanism reflects the process of how to urge all parties to take active measures through the feedback mechanism after the carbon emission status changes in the power industry. The R-D mechanism shows how the power industry can continuously generate new drivers through the virtuous circle mechanism after the implementation of the response measures to promote the development of the whole industry in the direction of low carbon and environmental protection. Based on the cycle mechanism of DSR model, combined with the characteristics of power industry and the actual demand of carbon reduction and emission reduction, an evaluation index system with 18 indicators is constructed to comprehensively evaluate the carbon reduction and emission reduction of power enterprises, as shown in Table 4.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Carbon audit evaluation indicators based on DSR model.

https://doi.org/10.1371/journal.pone.0316537.t004

Data mining is a process of discovering hidden information in large datasets through algorithms, which involves many technologies such as pattern recognition, machine learning and statistical analysis to support problem discovery in specific applications. In this study, machine learning algorithms such as decision tree method, regression analysis and cluster analysis are selected to mine and analyze the carbon emission data of power enterprises. Specifically, decision tree algorithm, as a supervised learning technology, is used to deal with classification and regression problems. Microsoft decision tree method provided by Microsoft SQL Server Analysis Services is used, which is suitable for forecasting and modeling of discrete and continuous attributes. When constructing the decision tree model, each node represents the test conditions of feature attributes, and each branch corresponds to the results within the range of feature attribute values. According to the characteristics of the dataset, the depth of the tree is adjusted (the maximum is set to 15), and the minimum sample number of leaf nodes is set to 5 to balance the complexity and accuracy of the model. At the same time, Gini index is used as the classification standard to optimize the tree construction process. In order to prevent over-fitting, pruning technology is applied and the minimum information gain threshold (0.01) is set.

In the construction process, the features that contribute the most to the prediction results are screened by calculating the importance of features, and the performance of the model is verified by using the 10-fold cross-validation method, thus improving the robustness and accuracy of the model. In data preprocessing, firstly, data cleaning, as the first step of preprocessing, aims to remove noise and inconsistent data in the dataset. Through statistical analysis and visualization methods, abnormal values and unreasonable data are identified. For example, for abnormal data such as fuel consumption or power generation, IQR method is used to detect and eliminate outliers, and repeated detection algorithm is used to eliminate possible data redundancy. Secondly, the treatment of missing values is also very important, because its existence may affect the accuracy and robustness of the model. The missing values of the data are analyzed, and corresponding processing methods are adopted according to the specific conditions of different characteristics. For key features (such as energy consumption data), interpolation method is used to fill in missing values to ensure the continuity and accuracy of data. For a small number of missing non-critical features (such as some equipment status data), the mean filling method is adopted. In addition, for features with high missing ratio, people choose to delete these features directly to avoid introducing too much uncertainty. Finally, after completing data cleaning and missing value processing, the data is standardized. Because the dimensions and ranges of different features are different, in order to prevent some features from having too great influence on the results in model training, Min-Max Normalization method is adopted to transform all eigenvalues into the range of [0,1]. At the same time, for the characteristics of normal distribution, Z-score standardization method is also used to ensure the uniformity of data and the effectiveness of analysis.

Through the above steps, it has laid a solid foundation for the subsequent data mining. In addition, in the application of Transformer architecture, a multi-layer encoder-decoder model based on self-attention mechanism is constructed to process the sequence data in carbon emission data. The parameter setting of this model includes the number of self-attention layers is 6, and the learning rate is 0.001. Adam optimizer is used for training, which aims to capture the long-range dependence in data and enhance the recognition ability of complex data patterns. Through the application of these algorithms, the hidden information in carbon emission data is successfully mined, which provides valuable knowledge support for power enterprises. In addition, based on the DSR model proposed by the United Nations Commission on Sustainable Development and the results of machine learning algorithm, this study constructs a carbon audit evaluation system consisting of 18 indicators to effectively evaluate and optimize the carbon emission management of the power industry. In the process of data analysis, regression analysis and cluster analysis are used to ensure the scientificity and integrity of the evaluation system. In the regression analysis, multiple linear regression and LASSO regression (minimum absolute contraction and selection operator) are used to establish the relationship between carbon emission data and various evaluation indexes, and the value of regularization parameter λ is set to 0.1 to balance the model complexity and prediction accuracy. By analyzing the coefficients of the regression model, the key factors affecting carbon emissions are determined, which provides targeted suggestions for enterprises’ carbon emission reduction strategies. In the aspect of clustering analysis, the carbon emission data of power enterprises are classified by K-means algorithm, and the number of clusters K is set to 3, and the best number of clusters is selected by elbow rule and contour coefficient analysis. This analysis makes it possible to identify enterprises that have outstanding performance in carbon emission reduction and those that need improvement, thus providing a basis for further carbon management strategy optimization.

There are significant differences in energy structure, economic development level and environmental policies in different regions, which challenges the effectiveness of carbon footprint management model. In some developing countries, such as South Africa and India, fossil fuels are still the main energy source. Therefore, carbon emission control strategies should focus on improving the efficiency of fossil fuels and promoting clean energy substitution. In contrast, in Europe and other places, with the progress of renewable energy technology, the policy focus has shifted to promoting the large-scale application of renewable energy. Therefore, the proposed carbon footprint management model has certain flexibility, and can adapt to the specific conditions of different regions by adjusting the input parameters to achieve more accurate carbon emission prediction and management. When building a carbon footprint calculation model based on big data and AI technology, this study pays special attention to collecting and processing a variety of key data, including but not limited to fuel types (such as coal, natural gas, nuclear energy and renewable energy), power generation, power generation efficiency, equipment operation status and geographical location. The accurate acquisition and processing of these data is the basis to ensure the accuracy and reliability of the model. For coal-fired power plants, the proportion of different quality coal and its corresponding carbon emission coefficient are recorded in detail. For wind power station, the influence of wind speed change on power generation efficiency is considered.

Raw data standardization processing

This study uses the carbon audit evaluation index system of the power industry, constructed in Table 4, along with the comprehensive evaluation method of entropy weight—TOPSIS to explore measures to ensure the effective application of the carbon audit evaluation index system. The study uses two power companies, China Resources Power and Datang International Power Generation Co., Ltd., as examples. It analyzes the effectiveness of carbon reduction and emission reduction in detail. Through the flush database and the official website of each enterprise, the original data from 2020 to 2022 needed to build the index system are obtained. Because each evaluation index’s nature and unit may vary, dimensionless original data must be used while building the carbon audit assessment index system for the power sector. Figs 5 and 6 display the two enterprises’ standardized matrices.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Data standardization results of China Resources Power enterprise.

https://doi.org/10.1371/journal.pone.0316537.g005

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Data standardization results of Datang International Power Generation Co., Ltd.

https://doi.org/10.1371/journal.pone.0316537.g006

Comprehensive evaluation of carbon governance in electric power enterprises

Figs 7 and 8 show the comprehensive evaluation results of China Resources Power and Datang International Power Generation Co., Ltd. from 2020 to 2022, and the index statistical results of each dimension evaluation. It shows that the comprehensive evaluation index of China Resources Power rose from 0.5458 in 2020 to 0.627 in 2022. This is mainly due to a series of positive measures taken by the company in clean energy investment, such as vigorously developing hydropower, wind power and photovoltaic projects, among which the construction of Jiangsu Lianshui wind farm has significantly reduced carbon emissions. China Resources Power gradually reduced the proportion of thermal power generation by vigorously promoting clean energy, such as hydropower, wind power and photovoltaic, thus significantly reducing carbon emissions. For example, Jiangsu Lianshui Wind Farm built by this enterprise has an annual power generation of 1.19 billion kWh, which can replace standard coal consumption by 11,700 tons and reduce carbon dioxide emissions by 26,000 tons. In addition, China Resources Power also actively invests in new energy projects, such as Chibi million-kilowatt new energy base, fishery and light complementary power generation project, etc. These projects not only provide a lot of green energy, but also significantly reduce carbon dioxide emissions. The comprehensive evaluation index of Datang International Power Generation Co., Ltd. is fluctuating, and it dropped to 0.4658 in 2021. It reflects the challenges it faces in carbon emission reduction. Especially in the response dimension, the company’s poor performance in sulfur dioxide (SO₂) and carbon dioxide (CO₂) emission reduction led to a significant decline in the evaluation index. This phenomenon can be partly attributed to changes in the external policy environment, such as the adjustment of emission reduction policies in some regions and the impact of power market fluctuations on the carbon emission costs of enterprises. In addition, the management strategies of enterprises in equipment maintenance and investment in environmental protection facilities may also be insufficient, resulting in poor performance in emission reduction. State dimension indicators have performed well in recent years, and the evaluation indexes have all reached above 0.7. In the response dimension, enterprises have encountered significant challenges. Especially in 2021, the evaluation index of this dimension dropped to 0.1955, when analyzing the potential influencing factors, the successful experience of China China Resources Power mainly benefits from its large-scale investment in clean energy projects, such as Lianshui wind farm built in Jiangsu and the million-kilowatt new energy base in Chibi, Hubei. These projects not only provide a lot of green energy for the company, but also significantly reduce coal consumption and carbon dioxide emissions. However, the volatility performance of Datang International Power Generation Co., Ltd. reflects its lack of flexibility and adaptability in response to market and policy changes, especially in 2021, its strategy adjustment in SO₂ and CO₂ emission reduction is not in place. After in-depth analysis of the reasons, it is considered that China Electric Power Resources Company can effectively reduce carbon emissions while the evaluation index is rising, mainly because of its strategic investment in clean energy, which gradually reduces the proportion of thermal power and increases the proportion of wind power and photovoltaic power generation, thus effectively reducing carbon emissions. In contrast, the progress of Datang International Power Generation Co., Ltd. is relatively slow, and the implementation effect of its carbon emission reduction measures failed to meet expectations, which may be due to its dependence on existing technologies and its untimely response to environmental policies. Based on the above analysis results, specific management suggestions are put forward for the two companies. For China Electric Power Resources Company, it is suggested to continue to strengthen investment in clean energy projects, further optimize the existing energy structure, and improve the real-time and accuracy of carbon emission monitoring to ensure the effectiveness of emission reduction measures. For Datang International Power Generation Co., Ltd., people should focus on improving the investment in environmental protection technology and strengthening the rapid response to policy changes, especially in the field of sulfur dioxide and carbon dioxide emission reduction technology. It is suggested to speed up the technology update and management optimization to ensure that the expected emission reduction targets can be achieved.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Index statistical results of comprehensive evaluation of China Resources Power and evaluation of various dimensions.

https://doi.org/10.1371/journal.pone.0316537.g007

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Index statistical results of Datang International Power Generation Co., Ltd. comprehensive evaluation and various dimensions evaluation.

https://doi.org/10.1371/journal.pone.0316537.g008

Table 5 shows the data comparison results before and after the improvement of carbon audit evaluation system of power enterprises in China and other countries:

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Data comparison results before and after the improvement of carbon audit evaluation system of power enterprises in China and other countries.

https://doi.org/10.1371/journal.pone.0316537.t005

Table 6 shows the changes of carbon audit evaluation indexes of China’s electric power enterprises (China Electric Power Company and Datang International Electric Power Company):

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Changes of carbon audit evaluation indexes of China’s electric power enterprises (China Electric Power Company and Datang International Electric Power Company).

https://doi.org/10.1371/journal.pone.0316537.t006

Under the background that the current carbon audit evaluation system and standards are not perfect, it is suggested to further improve these systems and standards in combination with the actual situation and advanced experience at home and abroad. By analyzing the data in Tables 5 and 6, the improvement of the system and standards has significantly improved the performance and effect of carbon audit in China’s power enterprises. According to the data in Table 5, before the improvement, the total carbon emission of China’s power enterprises is 50 million tons, the accuracy of carbon audit is 70%, the investment in low-carbon technology is 2 billion yuan, the carbon emission intensity is 0.8 tons/GWh, and the proportion of clean energy is 25%. After the improvement, these indicators have been significantly improved. The total carbon emission decreased to 40 million tons, the accuracy of carbon audit increases to 85%, the investment in low-carbon technology increases to 5 billion yuan, the carbon emission intensity decreases to 0.6 tons/gwh, and the proportion of clean energy increases to 35%. These data show that the improved carbon audit evaluation system has made remarkable progress in carbon emission control and low-carbon technology investment in China’s power enterprises, which reflects the effectiveness of the system improvement. Further analysis of the data in Table 6 shows the changes of carbon audit evaluation indexes of China Electric Power Company and Datang International Electric Power Company between 2020 and 2022. The overall evaluation index of China Electric Power has increased from 0.5458 in 2020 to 0.627 in 2022, the carbon emission reduction has increased from 2 million tons to 3.5 million tons, and the installed capacity of clean energy has increased from 50 million kilowatts to 70 million kilowatts. These data show that after the implementation of improvement measures, China Electric Power Company has significantly improved the accuracy and effect of carbon audit, and made remarkable progress in carbon emission reduction and clean energy investment. In contrast, the overall evaluation index of Datang International Power Company has decreased from 0.5237 to 0.4958. Although the accuracy of its carbon audit has also improved, the reduction of carbon emissions and the installed capacity of clean energy have changed little. This situation shows that although Datang International has improved the accuracy of carbon audit, it faces certain challenges in actual carbon emission reduction and clean energy investment. These data support the conclusion that there is still room for improvement in the current carbon audit evaluation system and standards.

By further optimizing the carbon audit system and combining the advanced experience in China and other countries, people can better meet the actual needs of power enterprises and improve the accuracy and effectiveness of carbon audit. The improved system can not only evaluate carbon emissions more accurately, but also promote enterprises to make greater progress in low-carbon technology investment and clean energy transformation, thus promoting the green development of the whole industry. Specifically, firstly, it is necessary to deeply analyze the operating characteristics of power enterprises and the specific links of carbon emissions, and clarify the key indicators of carbon audit evaluation. These indicators should cover all aspects of power production, transmission and distribution, consumption and so on to ensure that the evaluation system can fully capture the carbon emissions of enterprises. For example, indicators such as carbon emission intensity per unit power generation, clean energy ratio, carbon capture and storage capacity can be considered. Secondly, it draws on advanced theoretical frameworks such as the DSR model to construct a structured carbon audit evaluation system. DSR model provides a systematic analysis framework for carbon emission management by identifying drivers, evaluating current state and analyzing response measures. In the power industry, the drivers may include energy structure, technological progress and policy environment. The state involves carbon emissions, energy efficiency and other indicators. The response measures include the development of clean energy and the application of energy saving and emission reduction technologies. Furthermore, people should make full use of international authoritative data sources such as IEA CO₂ Emissions Database and refer to their carbon emission factors and accounting methods to formulate unified and standardized carbon footprint accounting standards. This not only helps to improve the accuracy of carbon audit, but also enhances the comparability of carbon emission data between different enterprises. In addition, a set of transparent reporting guidelines should be established to clarify the collection, calculation and reporting process of carbon emission data, as well as the third-party audit and verification mechanism. This will help to improve the transparency and credibility of carbon audit and promote the trust among enterprises, government and the public. Finally, the improvement of the evaluation system and standards should be a dynamic process, which needs to be updated and adjusted regularly according to the development trend of the industry, technological progress and policy changes. At the same time, experts and scholars, business representatives and policy makers inside and outside the industry are encouraged to participate in the construction and improvement of the evaluation system to ensure that the evaluation system is scientific, practical and forward-looking.