Influencing the Variable Selection and Prediction of Carbon Emissions in China

Abstract

In order to study the changing rule of carbon dioxide emissions in China, this paper systematically focused on their current situation, influencing factors, and future trends. Firstly, the current situations of global carbon dioxide emissions and China’s carbon dioxide emissions were presented via a visualization method and their characteristics were analyzed; secondly, the random forest regression model was used to screen the main factors affecting China’s carbon emissions. Considering the different aspects of carbon emissions, 29 influencing factors were determined and 6 main influencing factors were determined according to the results of the random forest regression model. Then, a prediction model for carbon emissions in China was established. The BP neural network model, multi-factor LSTM time series model, and CNN-LSTM model were compared on the test set and all of them passed the test. However, the goodness of fit of the CNN-LSTM model was about 0.01~0.02 higher than the other two models and the MAE and RMSE of the CNN-LSTM model were about 0.01~0.03 lower than those of the other two models. Thus, it was selected to predict China’s carbon dioxide emissions. The predicted results showed that the peak of China’s carbon emissions will be around 2027 and the peak of these emissions will be between 12.9 billion tons and 13.2 billion tons. Overall, the paper puts forward reasonable suggestions for China’s low-carbon development and provides a reference for an adjustment plan of energy structure.

1. Introduction

With the development of the global economy, carbon emissions continue to increase, leading to global warming and the destruction of the balance of nature. In order to strike a balance between economic development and the ecological environment, all countries are endeavoring to develop a “low-carbon economy”. As a major economic country in the world, China is paying more attention to its carbon emissions. In January 2022, the Political Bureau of the Central Committee of the Communist Party of China held its 36th collective learning session on “dual carbon”. The general secretary at the meeting made clear that “carbon peak and carbon neutrality” are inevitable requirements for promoting high-quality development and realizing green and low-carbon development in China. Therefore, the time to reach the peak of carbon and the peak of carbon emissions have become important research issues. At the same time, in order to develop practical green and low-carbon development programs, it is even more necessary to determine the main factors that affect carbon emissions.

The Japanese scholar Yoichi Kaya proposed that the Kaya identity is currently the main analysis method for analyzing the factors affecting carbon dioxide emissions, but it also has some shortcomings. Yuan Lu and Pan Jiahua specifically summarized three limitations of Kaya’s identity: firstly, it is impossible to explain the changes in stock; secondly, the actual impact of influencing factors on the total amount of carbon dioxide emissions is difficult to measure; and thirdly, the policy recommendations derived from Kaya identity are ambiguous and irrational, so they need to be analyzed and tested in conjunction with other factors [1]. Liu Jinhua extended the Kaya identity based on the LMDI decomposition model to decompose the factors affecting carbon emissions and effectively avoid the impact of data zero value [2]. Alexander V proposed the generalized Divisia index decomposition method [3]. This method not only overcomes the difficulty in removing the residual term of Kaya identity, but also compensates for the problem of factor decompositions covering each other in the LMDI method. Moreover, its decomposition results can distinguish the correlation between all factors, thus avoiding the problem of repeated calculations; Wang Feng et al. used the Logarithmic mean Divisia index decomposition method to decompose the growth rate of carbon dioxide emissions from China’s energy consumption into the weight of 11 influencing factors [4]. In addition, researchers have studied carbon emissions in different areas and regions. For example, Pan Xiaomei et al. studied the distribution characteristics and dynamic evolution process of China’s carbon emissions based on the barycentric model, demonstrating the correlation between economic development and carbon emissions [5]. Cheng Yu et al. applied the Super-SBM and STIRPAT model to investigate the spatial–temporal evolution of China’s carbon emission performance [6].

Energy consumption is one of the current key research issues and has a significant impact on the emissions of greenhouse gases. It poses a serious threat to the survival and development of human society. In order to achieve sustainable economic development under the dual constraints of energy consumption and environmental pollution, the government strongly supports technological innovation and the transition from a fossil-fuel-led energy system to a more sustainable low-carbon energy system [7]. Therefore, more scholars are investing in research on energy transformation and energy innovation [8,9].

The existing literature provides certain comprehensive theoretical and practical guidance for subsequent research, but it has some room for expansion. In terms of method application, most studies on the influencing factors of carbon dioxide are based on exponential decomposition models. Although this operation has the advantages of a simple calculation process and intuitive decomposition results, some important factors are difficult to include in the research scope due to the limitations of the models themselves; in terms of content, most studies focus on the influencing factors and evolution characteristics of carbon emissions. The research on future development trends and how to achieve green transformation still needs to be enriched and there needs to be more focus on the research about the influence factors of carbon dioxide emissions with less comprehensive research on them. Therefore, aimed at the problem of carbon dioxide emissions in China, this paper conducts a systematic and comprehensive study on the optimization of the model from its current situation, influencing factors, future development trends, and how to realize green low-carbon transformation in the future, thus putting forward reasonable suggestions for China’s green low-carbon development.

2. Data Collection and Processing

2.1. Data Source

Firstly, to study the current status of carbon dioxide emissions in the world, this article selected data on world economic development and carbon emissions from 1990 to 2020; in order to show China’s carbon dioxide emissions more intuitively, the carbon dioxide emissions by China and other countries were compared and the time span was selected to be from 1990 to 2019. All the data were from the World Bank (https://data.worldbank.org.cn, accessed on 18 April 2020), which uses carbon emissions data from the Carbon Dioxide Information Analysis Center of the Department of Environmental Science of the Oak Ridge National Laboratory in Tennessee.

Secondly, the influencing factors and prediction of carbon dioxide emissions were studied. From different perspectives of influencing carbon emissions in China, 29 variables, such as GDP, total population in China, energy intensity, private car ownership, and total length of transportation lines, were determined. The annual data of the above 29 variables from 2000 to 2020 were collected by consulting the China Statistical Yearbook, China Energy Statistical Yearbook, and the annual statistical bulletin of each province and city from 2000 to 2021.

2.2. Data Processing

Firstly, the interpolation method of multiple interpolation was used to process the missing data [10]. Multiple imputation is a method based on multiple simulations that can effectively solve the problem of missing data. This method uses the Monte Carlo algorithm to integrate all data sets containing missing data to obtain a complete data set. First, a value is selected that appears most frequently in a variable and the missing data are interpolate; then, the similarity between each observation value is calculated and iterated, and the similarity is calculated again on this basis, and interpolated again. This cycle ultimately forms five complete datasets.

As shown in Figure 1, the blue color in the imputed data graph represents the observed data, and the red color represents the imputed data. The results indicate that the distribution of the interpolation values and observation values was similar, and the interpolation effect was good. Therefore, the results of the fourth interpolation were made selected to form a complete data set.

Secondly, the data were standardized. On the one hand, the order of magnitude of each variable was different, and excessive numerical differences can cause bias in the research results; on the other hand, differences in the units between various feature variables and different dimensions can also cause errors in the research results. In order to eliminate the differences between the features and ensure that different features had the same scale, the method of Formula (1) was adopted to standardize the data.

$$\frac{x-\mu }{\sigma }$$

(1)

3. Current Situation of Carbon Dioxide Emissions in the World and China

At present, the development of the world today is a coexistence of opportunities and challenges. The global economy is gradually recovering in the post-COVID-19 era, and environmental problems brought about by economic development cannot be ignored. Nowadays, the problem of carbon dioxide emissions has attracted wide attention from all over the world. At the 75th United Nations General Assembly organized by the United Nations, various countries actively participated in the research on carbon dioxide emissions.

3.1. Current Situation of Global Carbon Dioxide Emissions

In general, carbon dioxide emissions show a rising trend, but the growth rate of these carbon dioxide emissions has gradually slowed down in recent years. Carbon dioxide emissions reached their current maximum value in 2018. Faced with increasing carbon dioxide emissions and increasingly serious global climate issues, the Paris Agreement has put forward a clearer vision of carbon emissions and made unified arrangements for coping with global climate change after 2020. From the perspective of a total of 178 countries around the world, this conference fully reflects that reducing carbon emissions has become a global issue and a consensus has been gradually formed around the world to deal with climate change. Therefore, with the development of major carbon-emitting economies and the implementation of carbon emission control policies and measures, the growth rate of global carbon emissions slowed down in 2010 and the growth rate of world carbon emissions fell rapidly after 2018, making the growth rate of world carbon emissions close to zero in 2019.

In addition, as can been seen from Figure 2 that carbon dioxide and economic development show a synchronous rising trend. With economic development, the demand for energy in various sectors continues to increase, and the consumption of energy is accompanied by a large number of carbon dioxide emissions, which makes the speed of economic development and carbon emissions accelerate at the same time. During a recession, energy consumption and carbon emissions both decline in stages. For example, the US subprime mortgage crisis in 2008–2009 and the COVID-19 pandemic in 2020 brought about zero growth or even a downward trend in total carbon emissions. In the economic recovery after the crisis, there will be a high recovery in carbon dioxide emissions. However, with the development of technology, a large amount of clean energy is gradually replacing fossil energy in people’s lives, and the production and utilization rate of energy has also significantly improved. Therefore the relationship between economic development and carbon emissions tends to weaken.

3.2. Current Situation of Carbon Dioxide Emissions in China

With the rapid development of China’s economy and the continuous expansion of its industrial production scale, the emissions of carbon dioxide are also increasing. The trend of China’s economic development and carbon dioxide emissions is shown in Figure 3.

Similar to the global trend of carbon dioxide emissions, China’s carbon emissions have also shown a synchronous upward trend with economic growth. Especially during 2000–2010, there was an obvious positive correlation between China’s carbon dioxide emissions and GDP growth, and the growth rate gradually increased. Since 2011, China has been paying attention to environmental protection issues. The implementation of a series of environmental policies and the use of low-carbon energy have slowed down the growth rate of carbon dioxide emissions.

In recent years, China’s carbon dioxide emissions have maintained a growing trend compared to some other countries. The carbon emissions of other countries are mostly in a relatively stable state and may even show a downward trend. As shown in Figure 4, China’s carbon dioxide emissions surpassed those of the United States in 2005, becoming the world’s largest carbon dioxide emitter with emissions far higher than other countries [11]. However, since 2010, the growth rate of carbon dioxide emissions in China has decreased. India is another country besides China that is gradually increasing its carbon dioxide emissions.

Figure 5 shows the percentage differences in the carbon emissions among several other countries, with the United States as a reference. The results confirm that there are significant differences in the carbon emission changes in China, especially between 2000 and 2010. In the end, it not only surpassed the United States, but also more than doubled its size. Several other countries are in a relatively stable state.

There are many reasons why China’s carbon emission trends differ from those of other countries. At present, China is currently in an important period of socialist modernization construction, so carbon dioxide emissions are an inevitable problem. Moreover, China is a major trading country, and trade processing accounts for a large proportion of these emissions. Thus, the rapid growth of China’s trade is at the cost of a lot of pollution and carbon emissions. However, China also recognizes the importance of environmental protection, and therefore strongly supports the implementation of environmental policies. Therefore, the growth rate of carbon dioxide emissions in China has slowed down.

Although the total amount of carbon dioxide emissions continues to increase in China’s modernization process, China’s per capita carbon dioxide emissions are lower than those of other countries.

Figure 6 shows that China’s per capita carbon dioxide emissions increased significantly in 2003–2012. The carbon dioxide emissions per capita in some developed countries have been relatively stable or even declining. Overall, China’s per capita carbon dioxide emissions are lower than those of most countries in the world. As various countries have considered climate change as a serious threat, China has also begun to take a series of measures to control its carbon dioxide emissions. After 2011, China’s per capita carbon dioxide emissions no longer showed obvious growth, and the development was relatively stable.

The above-mentioned countries are major industrial countries that heavily rely on the large-scale use of fossil fuels such as coal, oil, and natural gas. Travel is highly dependent on private cars, and the capacity of urban public transportation and interstate railway passenger transportation is insufficient. This has also resulted in a large amount of carbon dioxide emissions. China is the country with the largest population in the world. Its environmental awareness is constantly increasing with the country’s emphasis on green development. Therefore, its per capita carbon emissions are lower than those of other countries in the world.

4. Variable Selection of Influencing Factors of Carbon Dioxide in China

Research in different fields and industries has found that the factors affecting carbon dioxide emissions are different. Economic development level, population size, transportation infrastructure, industrial structure, eco-investing, and environmental regulation have significant impacts on the carbon emissions of the circulation industry [12]; energy structure, energy intensity, and economic level on the emissions of the construction industry; population density, rate of new energy-saving building areas, and new building areas have an impact on the carbon emissions of the construction industry [13]; and the sown area of crops, number of large-scale livestock, rural per capita disposable income, agricultural mechanization level, and urbanization rate affect agricultural carbon emissions [14]. There are many factors affecting the carbon dioxide emissions from various perspectives and 29 variables are summarized in this paper. However, due to its own limitations, the exponential decomposition model makes it difficult to include various influencing factors in the scope of the research. The random forests regression model can realize the measurement of the importance of the characteristics of multiple variables, so this article is based on the 29-variable regression model for variable selection by random forests.

The random forest regression model is composed of multiple regression trees [15]. There is no correlation between the decision trees. The final result is determined by each decision tree in the whole forest. For each decision tree, the corresponding out-of-bag (OOB) data are selected to calculate the out-of-bag data error, denoted as the OOB error. In the process of each decision tree construction, a set of decision tree data is obtained through bootstrap resampling technology for training, while the approximately one third of the remaining data is not utilized and referred to as out-of-bag data. These data can be used to evaluate the performance of the decision trees. For each decision tree, its out-of-bag data error is calculated, denoted as the OOB error_0. Another feature is selected, the sequence is changed randomly, and the out-of-bag data error is calculated again, denoted as the OOB error_1. The importance of a variable can be estimated by analyzing the increase in out-of-bag data error when the data series outside the bag is changed, as shown in Equation (2):

$$\frac{{\displaystyle \sum (\mathrm{oob} \mathrm{error}\_1-\mathrm{oob} \mathrm{error}\_0)}}{N}$$

(2)

The reason why this value can demonstrate the importance of a feature is that, if the accuracy of the calculated out-of-bag data decreases significantly after randomly changing the sequence, it indicates that this feature has a significant impact on the prediction results of the sample, and thus indicates that this feature is of a greater importance to the dependent variable.

Firstly, 29 factors affecting carbon dioxide emissions, such as the total population, urbanization rate, and the total length of transportation lines, were selected and standardized to establish a random forest regression model. The total data set was divided into a training set (70%) and a test set (30%). The training results showed that nine variables were randomly selected from the nodes of each tree to generate 500 decision trees, so that the errors tended to balance as shown in Figure 7.

The mean squared error of the finally trained model in the training set was 0.019, the mean squared error of the test set was 0.052, and the goodness of fit was more than 90% (see

Table 1. Training results of decision tree error model.

	MSE	R2
Train set	0.019	0.978
Test set	0.052	0.948

R²: goodness of fit. MSE: mean squared error.

), indicating that the model had a good performance. Then, the selected variables were measured using the above method for feature importance. The results are shown in Figure 8.

According to the return results of the random forest regression model, the 29 variables were summarized and eliminated. Finally, 15 variables were selected and summarized into 6 main factors affecting carbon dioxide emissions, which were energy utilization, industrial structure, energy structure, transportation, population scale, and economic development. The specific indicators are shown in

Table 2. Influencing factors of carbon dioxide emission.

Basic Index	Specific Index
Energy utilization	Intensity of energy
	Processing in energy efficiency
Industrial structure	Proportion of investment in primary industry
	Proportion of investment in the secondary industry
	Proportion of investment in tertiary industry
	Proportion of fixed assets investment in the whole society
Energy structure	Raw coal production
	Crude oil production
	Diesel consumption
	Coke consumption
Transportation	Number of civilian car owned
	Number of vehicles in highway operation
	Number of civilian ships owned
Population scale	Total population
Economic development	Gross domestic product

.

5. Predic p5 Prediction of Carbon Dioxide Emissions in China

5.1. Construction of a Carbon Dioxide Prediction Model

The goals of “carbon peak” and “carbon neutrality” not only provide guidance for the comprehensive green transformation of China’s economic and social development, but also make important contributions to the global response to climate change. Obviously, the time of the carbon peak has a direct impact on the difficulty of achieving China’s dual-carbon goal. To develop practical and feasible solutions for green, low-carbon, and sustainable development, it is necessary to predict the peak and timing of carbon dioxide emissions. Therefore, this paper predicts the situation of the carbon peak in China.

Neural network models are widely used in various fields such as prediction, pattern recognition, and risk assessment [16,17,18]. The BP neural network model is a multi-layer feedforward network trained according to error backpropagation [19]. It belongs to a supervised learning algorithm. A classic BP neural network consists of three layers: an input layer, one or more hidden layers, and an output layer. The neurons in the input layer are responsible for receiving important information about various factors affecting carbon emissions and passing this information to the neurons in the middle layer. The output layer outputs the predicted value for future carbon emissions after information processing to the outside world. The hidden layer is the internal information processing layer, which realizes the transformation of information through continuous updating and iteration. Traditional prediction methods are not suitable for situations where there are a lack of historical cost data and many influencing factors. However, it also has some shortcomings which require a large amount of data for training and learning; otherwise, it is prone to overfitting or underfitting problems.

Most of the time, only time and the variables themselves can be used for reference in the established model, but there are often many other factors besides time that will affect the variables. Traditional linear prediction methods are difficult to apply in the case of multiple variables and multiple inputs. Long short-term memory (LSTM) is a special recurrent neural network (RNN) that can solve the problem of multiple input variables well [20]. The LSTM model has a wide range of applications in different fields, such as finance, geology, meteorology, and so on [21,22,23,24]. It introduces a selective mechanism of “gating” based on RNN to selectively retain or delete information in order to better learn long-term dependencies. Changes in variables at any time in the past will have an impact on the trend of carbon emissions, and the longer the time node is, the smaller the impact will have. Therefore, the gating mechanism of LSTM can fully play its role in filtering out less influential information and filtering out information from historical data that is effective for future prediction to improve the prediction accuracy. However, there are some limitations when the time span is large and the calculation takes time.

Convolutional neural networks, CNNs, mainly consist of a convolution layer, linear rectification layer, pooling layer, full connection layer, and other structures. Among them, convolutional layers play a very significant role in extracting different input features. As the number of convolutional layers increases, more and more networks are able to iterate from low-level features to obtain more information about high-level features. The CNN model also has shortcomings. Using the CNN model alone, it can be found that the convolution operation of each period of time series data has a good performance in extracting the local features of carbon dioxide emissions. However, CNNs are not sensitive to the temporal sequence information related to carbon dioxide emissions and cannot perform prediction tasks independently. The LSTM model has certain limitations. In the process of using the LSTM model, it can be found that noise will be introduced, which has nothing to do with the prediction of carbon dioxide emissions. At the same time, large and small values in the data will affect the robustness of the model, leading to a large error in the prediction effect compared to the actual. Therefore, a new model (called CNN-LSTM model) is built by combining the advantages of extracting CNN and LSTM, respectively, which makes full use of CNN’s feature information extraction capability and LSTM’s sensitivity to time series data [25,26,27,28], thus improving the prediction of carbon dioxide emissions.

The construction of a prediction model and parameter optimization are conducted. By training the training set, the BP neural network model chooses to build two hidden layers and sets the number of neurons in the hidden layer to 10. An LSTM neural network is built using Keras to add a dropout layer during the model training in order to prevent fitting problems; a CNN LSTM model is established to extract the information that needs to be retained from the data by constructing two one-dimensional convolutional layers and two maximum pooling layers at the CNN through model training; since the data in this paper contain time series, two LSTM layers are finally constructed through training.

The loss function of the final BP neural network model training is shown in Figure 9a. The analysis shows that the loss function of the model gradually decreases with the increase in the number of training iterations. The speed of convergence is faster before the number of epochs is 10. After the number of epochs is about 40, the curves of the training set and the test set almost coincide and the loss function gradually converges and becomes stable. According to Figure 9b, the loss function image error decreases rapidly at the beginning. When the number of epochs is about 20, the curves of the loss function of the training set and the test set almost coincide, indicating that the model is robust and the training is complete. The loss function image of the CNN-LSTM model is drawn as shown in Figure 9c. In the beginning, the loss function shows a trend of a rapid increase and sharp decline. However, with the increase in the number of iterations, the difference between the loss function of the test set and the training set is small when the number of epochs approaches 40, indicating that the model achieves a relatively ideal training effect.

The loss function graph shows that the model training achieves the expected training effect, but the model with a better forecasting effect cannot be selected. In order to evaluate the prediction ability and accuracy of the model, R² (goodness of fit), MAE (mean absolute error), and RMSE (root mean square error) are used for a comparative analysis of the prediction models. The calculation formula is as follows:

$$R^2=\frac{{\displaystyle \sum _{i=1}^n{({\widehat{y}}_i-\overline{y})}^2}}{{\displaystyle \sum _{i=1}^n{(y_i-\overline{y})}^2}}$$

(3)

$$MAE=\frac{1}{n}{\displaystyle \sum _{i=1}^n\left|y_i-{\widehat{y}}_i\right|}$$

(4)

$$RMSE=\sqrt{\frac{1}{n}{\displaystyle \sum _i{(y_i-{\widehat{y}}_i)}^2}}$$

(5)

The calculation results are shown in

Table 3. Model training errors.

	R2	MAE	RMSE
BP	0.9572	0.1726	0.2068
LSTM	0.9467	0.1901	0.2309
CNN-LSTM	0.9619	0.1629	0.1951

R²: goodness of fit. MAE: mean absolute error. RMSE: root mean square error.

.

Finally, the error results of the training of the three models show that the goodness of fit of the BP neural network model, the LSTM model, and the CNN-LSTM model are all around 0.95, which indicates that the fitting effect of the three models is good. In particular, the goodness of fit of the CNN-LSTM model is about 0.01~0.02 higher than the other two models, and the MAE and RMSE of the CNN-LSTM model are about 0.01~0.03 lower than those of the other two models. Therefore, the CNN-LSTM model is finally selected to predict future carbon emissions.

5.2. Results of Carbon Dioxide Prediction

A reliability test is carried out in order to further test the prediction effect of the CNN-LSTM carbon dioxide emissions prediction model. The index data of economic development, population size, and other factors that had a large influence on the carbon dioxide emissions from 2000 to 2020 are substituted into the model to analyze the trained CNN-LSTM carbon emissions prediction model. The visual results from comparing the predicted data with the actual measured carbon emissions data are shown in Figure 10.

The measured value has a high coincidence with the predicted value curve, and the historical trend is roughly the same. It is proved that the trained model has a good reliability and can be used to predict the future trend of carbon dioxide emissions in China based on the CNN-LSTM model.

Through the model training error test and reliability test, the trained CNN-LSTM model is finally selected to predict the situation of the carbon peak in China. The prediction results are shown in Figure 11.

The results confirm that China’s carbon emissions have shown a growing trend, and it is predicted that this growth will stop around 2027. Carbon emissions also peaked in the range from 12.9 billion to 13.2 billion tons. After that, they gradually decreased and eventually flattened out. Du Yan et al. predicted that China’s carbon peak time is between 2025 and 2030, and the peak value is between 11.1 billion tons and 13.0 billion tons [29]. Sun Meng et al. predicted that China could achieve different degrees of carbon peak in 2028 or 2029, with the peak carbon emissions being between 12 billion tons and 12.2 billion tons [30].

These results are very similar, but there are small differences. The reason may be the error of the prediction model’s accuracy or the influence of carbon sink and influencing factors. However, this is within acceptable limits.

6. Policy Suggestions

This article conducted a comprehensive study on the current situation, influencing factors, and future development trends of carbon dioxide emissions in the world and China. China is still in the stage of high-quality development and its economy still needs to continue to grow. The problem of carbon emissions is inevitable. The realization of the “double carbon” goal has certain great obstacles to overcome. Based on this situation, China’s green and low-carbon transformation strategy should focus on the following aspects:

• The promotion and development of low-carbon life. The country’s total population, number of civilian cars owned, and number of civilian ships owned are the main factors affecting the carbon emissions in China. From the perspective of residents’ consumption, China should vigorously promote a green and low-carbon lifestyle across the whole society. For example, in terms of transportation, China should develop and invent various forms of travel tools with a high quality to guide residents to travel green, increase energy efficiency, and reduce carbon emissions.
• The vigorous development of new energy sources. The consumption of fossil energy is still one of the most important sources of carbon dioxide. In order to achieve the “double carbon” goal and accelerate the popularization of low-carbon and even zero-carbon energy, low-carbon and zero-carbon energy should be included in the energy structure and their proportions should be increased, assisting energy transformation with technological innovation to achieve a qualitative leap in the efficiency and performance of photovoltaics.
• Industrial upgrades. In China, secondary industry is still the main factor affecting carbon emissions. Thus, China should continue to upgrade and optimize its industrial structure to help reduce carbon emissions. From the perspective of energy, there is space for continuous improvements in China’s energy intensity and energy efficiency, especially in the secondary industry of the economy, which can form a useful supplement in industrial upgrading and work together to reduce China’s carbon emissions.