Research on the potential for China to achieve carbon neutrality: A hybrid prediction model integrated with elman neural network and sparrow search algorithm

Abstract

China's carbon reduction is of substantial significance in combating global climate change. In the context of the COVID-19 epidemic hit and economic and social development uncertainty, this study intends to discover whether China can attain the strategic destination of carbon peaking by 2030 and carbon neutrality by 2060 on schedule. Toward this aim, the grey relation analysis (GRA) is applied to filter the elements influencing carbon emissions to downgrade the dimensionality of indicators. A hybrid prediction is proposed integrated with Elman neural network (ENN) and sparrow search algorithm (SSA) to explore the potential for China to carbon neutrality from 2020 to 2060. The results reveal eight elements including GDP per capita, population, urbanization, total energy consumption and others are highly correlated with carbon emissions. China has a good chance of carbon peaking from 2028 to 2030, with a value of 11568.6–12330.5 Mt, while only one scenario can achieve carbon neutrality in 2060. In the neutral scenario, China should reach a proportion of renewable energy exceeding 80%, the urbanization rate reaching 85% and energy consumption controlling within 6.5 billion tons. A set of countermeasures for carbon abatement are presented to facilitate the implementation of carbon neutrality strategy.

Nomenclature

GHG

greenhouse gas

Mt

million tons

GRA

grey relation analysis

ENN

elman neural network

SSA

sparrow search algorithm

BP

back propagation

LMDI

logarithmic mean Divisia index

SDA

structural decomposition analysis

CSO

chicken optimization algorithm

SVM

support vector machine

CCUS

carbon capture utilization and storage

IPAT

environmental impact, population, affluence, and technology level

l

sparrows' number

p

variable dimension

t

current number of iterations

$X_{i,j}^t$

the factor of the jth dimension of the ith sparrow at the tth iteration

${\text{iter}}_{\max }$

indicates the constant with the highest number of iterations

Q

a random number

k

training number

x

hidden layer neuron output vector

y

output vector

x_c

feedback state vector

u

input vector

g

transfer function of the output neuron

w

connection weight matrices

f

transfer function of hidden layer neuron

GDP

Gross domestic product

PGDP

GDP per capita

SSP

Proportion of secondary sector to GDP

FAI

Fixed asset investment

TSP

Proportion of tertiary sector to GDP

TEC

Total energy consumption

CCP

Proportion of coal consumption in primary energy

EI

Energy intensity

NFC

Proportion of Non-fossil energy in total energy consumption

FFP

Proportion of fossil fuels

EC

Electricity consumption

TPP

Proportion of thermal power

UR

Share of urban population in total

TP

Total population

OL

Openness level, share of import and export to GDP

RI

Proportion R&D expenditure in GDP

MAPE

mean absolute percent error

MAE

mean absolute error

RMSE

root mean squared error

1. Introduction

The issue regarding controlling greenhouse gas (GHG) emissions is a collective challenge globally, and how to mitigate carbon emissions effectively to curb the global warming trend has become a popular concern. Many countries are initiating to pursue carbon neutrality to confront global warming, i.e., to restrain GHG emissions into the atmosphere (Fan et al., 2022). The United Kingdom leads the way in proclaiming net zero carbon emissions by 2050. A close second is Sweden, France, and New Zealand (Fan et al., 2022). “The European Green Deal” (Malka et al., 2022) has achievable goals to mitigate the GHGs. Countries across the world seek net-zero emission pathways in sectors such as food (Becker et al., 2020) and cement (Sanjuán et al., 2020) and have started to evaluate carbon-neutral pathways for countries (Ji et al., 2021; Su et al., 2021; Tao et al., 2021) and specific regions (Kourgiozou et al., 2021). As the leading emitter, China produced 101.75 million tons (Mt) CO₂ in 2019, accounting for about 27.9% of global carbon emissions. To cope with this crucial challenge, the Chinese government plans to hit top emissions before 2030 and attain net zero emissions by 2060.

Nevertheless, it takes more than 60 years for developed countries, such as the UK, to drop from peak to net zero carbon emissions, while the time for China is merely 30 years. The heavy industry-dominated industrial structure and fossil fuel-dominated energy mix in China has exacerbated the dire situation of emission reduction. Concurrently, affected by the COVID-19 epidemic, China's economic development pace and production patterns have been hit heavily, making it uncertain whether China can attain its carbon neutrality by 2060. Hence, there are concerns about China's aggressive ambitions. This research endeavors to shed light on the factors shaping China's carbon emissions, and to construct an efficient carbon emission prediction model to uncover the evolutionary route of China's carbon emissions in the future, thereby investigating the potential of the country to hit its carbon neutrality targets on schedule. Furthermore, it needs further to clarify the socio-economic requirements to meet the carbon neutrality ambition. The solution to the issues above will offer powerful support and reference for the design of relevant policies and market mechanisms. Moreover, given those traditional models such as IPAT and LMDI can hardly handle high-dimensional data, and the prediction efficiency of single prediction methods still needs to be further improved, an efficient carbon emission prediction technique is urgent to be developed.

To achieve this purpose, this paper systematically summarizes and screens out the core drivers in terms of carbon emission, and by constructing a hybrid prediction model to detect the carbon emission development paths under different scenarios of development, the economic and social conditions for achieving carbon neutrality are revealed. The contributions are as follows: (1) Quantify the socio-economic considerations influencing carbon emissions. By introducing the gray relation analysis (GRA) algorithm to rank the correlations of the influencing parameters, eight core elements were extracted. Based on this, the above elements are used as input variables for the prediction model, which greatly reduces the dimensionality of the data for the subsequent prediction model and boosts the operational efficiency of the prediction model; (2) Propose a novel framework of carbon emission projection methods integrated with elman neural network (ENN) and sparrow search algorithm (SSA). Compared with the Back Propagation (BP) neural network and BP-SSA algorithm, the proposed model shows distinct superiority regarding the accuracy and operational efficiency of prediction. The framework enriches the practice fields of the algorithm and presents a benchmark for carbon emissions research; (3) Explore the carbon emission pathways under different economic and social energy system development scenarios, reveal the socio-economic conditions that need to be met for China to achieve carbon neutrality, and provide the basic guidelines for the direction of carbon reduction policies to be worked on.

The article is organized as follows: the literature review is presented in Section 2. A hybrid forecasting model GRA-ENN-SSA for carbon emissions is proposed in Section 3. Section 4 shows the results which contain the validation of the GRA-ENN-SSA model and analysis of model superiority, and an exploration of China's carbon-neutral pathway. Section 5 shows the conclusions.

2. Literature review

Scholars around the world have conducted a wealth of research regarding carbon emission-related issues. Currently, the general research on carbon emissions categorizes into the following two dimensions.

One category dedicates to seeking the drivers of carbon emissions via historical emission data. Many scholars employ the logarithmic mean Divisia index (LMDI) model, Kaya model, IPAT model, and structural decomposition analysis (SDA) as instruments to reveal the impact of changes on carbon emissions by analyzing scale factors (economy, population), structural factors (industrial structure), technological factors and social factors. Specifically, Li et al. (2019) employed the LMDI to uncover what's driving the change in carbon emissions, namely, carbon emission efficiency, economic growth, industrial structure, and energy intensity, and thereby forecast China's possible peaking scenario. Employing the SDA, Mi et al. (2017) found that drivers of carbon emission in China have shifted since 2008, and production structure and consumption structure have emerged as inhibitors of carbon emissions, whilst consumption levels have noticeably pushed emission growth. Helping with the Kaya model, Wang et al. (2019b) found that the impact of urbanization on the evolution of emissions may present more complex; the intensive urbanization process will allow the efficiency of carbon reduction to be enhanced. Urbanization may hinder the carbon peaking process due to the growing urban population, the increasing investment in infrastructure and the progress of consumption structure (Liu et al., 2020). Despite the literature above presenting an insight into factors influencing carbon emissions, it suffers an absence of research on the rising crucial factors like scientific and technological level and clean energy penetration.

The other category is about carbon emission projections within distinct sceneries. On one hand, many scholars apply conventional methods including the STIRPAT model (Fang et al., 2022; Sun et al., 2022b), system dynamics (Li et al., 2021), dynamic stochastic general equilibrium model (Yang et al., 2021a), and other quantitative models to exploit the carbon emissions. Numerous researchers claim that China will be likely to hit a carbon peak in 2022–2025 (Yu et al., 2018),2025–2030 (Wang et al., 2019a), 2029–2031 (Li et al., 2022a), 2032 (Niu et al., 2020), 2029–2035 (Xu et al., 2019) and 2028–2040 (Fang et al., 2019). Nevertheless, the results are inconsistent or even contradictory due to the differences in data, research methods, and other considerations (Fang et al., 2019; Jiang et al., 2019). Yang et al. (2021b) performed simulations and evaluated CO₂ emissions by adopting a multisectoral dynamic stochastic general equilibrium model. The results show energy restructuring could contribute to meeting the carbon reduction target in 2030, which opens the way up to carbon neutrality in 2060. Su and Lee (2020) discussed when China will reach the top via the STIRPAT model. The findings suggest China has the chance to reach a peak by 2028, with a value of 117.70 Mt. Li et al. (2021) proposed a system dynamics method to research the carbon emissions of the construction industry. Similar research was also performed by Zhang et al. (2022b). In addition, as artificial intelligence technology gradually matures and is applied, machine learning methods, such as chicken optimization algorithm (CSO) (Ren and Long, 2021), support vector machine (SVM) (AlKheder and Almusalam, 2022), BP neural network (Dong et al., 2018; Wen and Yuan, 2020), are widely used in carbon emission prediction because of their powerful performance. In addition, an effective combination of intelligent algorithms and artificial intelligence helps eliminate the drawbacks of a single approach and becomes an interesting area of research in forecasting. Wen and Cao (2020) proposed the SVM and CSO algorithm model to study residential CO₂ emissions in Shanghai. Qiao et al. (2020) proposed a new model by effectively integrating least squares SVM and intelligent algorithms and revealed carbon emissions for multiple countries. References (Li et al., 2022a, 2022b; Sun and Huang, 2022) used the Aquarius algorithm, the bat algorithm, and others to optimize the extreme learning machine, and constructed a hybrid prediction model to research the carbon neutralization path in China. However, technical countermeasures including carbon capture utilization and storage (CCUS) were not considered in the above research. As a swarm intelligence algorithm, the sparrow search algorithm (SSA) works by simulating the feeding behavior of sparrows (Li et al., 2022c). The SSA shows the merits of fast convergence speed and a small number of adjustment parameters compared to the GA, LSO and so on (Li et al., 2022e). As for the forecasting domain, Inefficient optimization usually indicates flawed results and a shortage of predictive capability (Khalid and Javaid, 2020). Fortunately, algorithms such as SSA are extensively utilized in optimization problems (Li et al., 2022c; Liu and Rodriguez, 2021; Sun et al., 2022a; Zhang et al., 2022a). Compared to other neural networks such as BP, ENN is the recurrent network and can effectively store the hidden neuron value of the former step, and more accurately identify data features. Hence, the ENN has been embraced by many researchers and fields. (Boualem et al., 2022; Ding et al., 2022; Meng et al., 2022; Ruiz et al., 2018).

Given the gap in existing works, this study overcomes the drawbacks of single optimization algorithms or linear models to forecast CO₂ emissions and proposes a prediction model to enhance the projection capability and accuracy. Moreover, carbon capture and utilization, and storage are considered, and nine different socioeconomic development sceneries are constructed to evaluate China's carbon neutrality path. This is particularly beneficial for policymakers to search for a scientific net-zero emission solution and clarify the implementation direction of emission reduction policies.

3. Methods and data

3.1. Factors influencing carbon emissions

IPAT method is one of the popular instruments to investigate energy economics and is also extensively employed in carbon emission forecasting, but the model suffers from the drawback that the independent variables that affect carbon emissions all displayed the same hierarchical relationship. York et al. (2003) extended the previous studies and proposed the STIRPAT model, that is:

$$I=a{P}^b{A}^c{T}^{d}e$$ (1)

where, I, P, A, and T denote environmental impact, population, affluence, and technology level, respectively; e is the error; a is the constant, b, c, and d are the parameters to be estimated.

To make a more accurate prediction outcome, literature (Niu et al., 2020; Ren and Long, 2021; Su and Lee, 2020) further extended the indicators related to carbon emissions via the STIRPAT model. Hence, in this work, indicators reflecting carbon emissions are obtained from economic, social, and energy systems based on the literature above, as shown in Table 1 .

Table 1.

The indicators related to China's CO₂ emissions.

Dimension	Factor
Economy	GDP, PGDP, SSP, FAI, TSP
Energy	TEC, CCP, EI, NFC, FFP, EC, TPP
Society	TP, UR, OL, RI

3.2. Gray relation analysis (GRA)

GRA is an analytical method that specifically addresses multifactorial and nonlinear problems. The correlation degree of each factor is judged depending on the similarity of their trends (Li et al., 2022d). The method can make up for the drawback that the mathematical and statistical methods have restrictive requirements on the sample. GRA is mainly divided into the following steps:

• (1)Identify the reference sequences that characterize the behavior of carbon emissions $X_0=x_0\left(k\right),k=\mathrm{1,2},...,n$ and the comparison sequences that influence the change in carbon emissions $X_i=x_i\left(k\right),i=\mathrm{1,2},...,m$.
• (2)Homogenization process: $x_i\left(k\right)=\frac{x_i\left(k\right)}{\overline{x_i}}$

where k is the time and i is a row in the subsequence.

• (3)Calculate the correlation coefficient (Li et al., 2022d), ${\Delta }_i\left(k\right)=\left|x_0\left(k\right)-x_i\left(k\right)\right|$.

$${\xi }_i\left(k\right)=\frac{\underset{i}{\mathit{min}}\underset{k}{\mathit{min}}{\Delta }_i\left(k\right)+\rho \underset{i}{\mathit{max}}\underset{k}{\mathit{max}}{\Delta }_i\left(k\right)}{{\Delta }_i\left(k\right)+\rho \underset{i}{\mathit{max}}\underset{k}{\mathit{max}}{\Delta }_i\left(k\right)}$$ (2)

where ρ is the resolution coefficient.

• (4)Calculating the degree of correlation

In this work, the average correlation of each stage was collected to reveal the relationship between the indicators. The correlation degree is calculated as Eq. (3).

$$r_i=\frac{1}{n}\sum _{k=1}^n{\xi }_i\left(k\right)$$ (3)

The correlations are ranked, if $r_1<r_2$, then the correlation of x₂ is higher.

3.3. Sparrow search algorithm (SSA)

The principle of SSA is derived from the predatory behavior of sparrows and SSA shows the characteristics of fast convergence and great optimization capability. The algorithm classifies the entire sparrow population into three categories, namely joiners grabbing for food, producers looking for food, and vigilantes detecting danger (Jia et al., 2022). The algorithm first needs to initialize the population with fitness value, and the sparrow population is initialized as followings:

$$X=\left(\begin{array}{cccc}{x}_{\mathrm{1,1}}& x_{\mathrm{1,2}}& ...& x_{1,p}\\ x_{\mathrm{2,1}}& x_{\mathrm{1,2}}& ...& x_{2,p}\\ ⋮& ⋮& & ⋮\\ x_{l,1}& x_{l,2}& ...& x_{l,p}\end{array}\right)$$ (4)

where l indicates the sparrows' number; p is the variable dimension; $x_{l,p}$ is the value of the pth dimension of the lth sparrow. The overall sparrow fitness value is characterized in the form:

$$F_X=\left(\begin{array}{c}f\left(\left[\begin{array}{cccc}{x}_{\mathrm{1,1}}& x_{\mathrm{1,2}}& ...& x_{1,p}\end{array}\right]\right)\\ f\left(\left[\begin{array}{cccc}{x}_{\mathrm{2,1}}& x_{\mathrm{2,2}}& ...& x_{2,p}\end{array}\right]\right)\\ ⋮\\ f\left(\left[\begin{array}{cccc}{x}_{l,1}& x_{l,2}& ...& x_{l,p}\end{array}\right]\right)\end{array}\right)$$ (5)

where f(x) is the individual fitness value. Sparrows with better fitness values (i.e., producers) will have priority in the search for food, while producers will have a larger foraging search range.

The producer's position constantly shifts during foraging, and the movement rule changes when it encounters a predator (Li et al., 2022c):

$$x_{i,j}^{t+1}=\{\begin{array}{cc}{x}_{i,j}^t\exp \left(\frac{-i}{\alpha {\text{iter}}_{\max }}\right),& R_2<ST\\ x_{i,j}^t+QL,& R_2\ge ST\end{array}$$ (6)

Among them, t indicates the current number of iterations; $j\in \left\{\mathrm{1,2},...,p\right\}$; $X_{i,j}^t$ indicates the factor of the jth dimension of the ith sparrow at the tth iteration; $\alpha \in (\mathrm{0,1}]$, indicates the random number; ${\text{iter}}_{\max }$ is the constant with the highest number of iterations; Q is a random number; $R_2\in \left[\mathrm{0,1}\right]$ is the alarm value; $ST\in \left[\mathrm{0,1}\right]$ is the safety threshold; L is a matrix.

As for the joiners, once the producer finds a good food source during the foraging process, the joiners will certainly know about it and fly to its vicinity to grab food, while there are also joiners who will always monitor the producer and are ready to compete for food. The rules for updating the position of the joiners are as follows (Li et al., 2022e)：

$$x_{i,j}^{t+1}=\{\begin{array}{cc}Q\exp \left(\frac{x_{worst}^t-x_{i,j}^t}{t^2}\right),& i>\frac{n}{2}\\ x_p^{t+1}+\left|x_{i,j}^t-X_p^{t+1}\right|A^{+}L,& otherwise\end{array}$$ (7)

$$A^{+}=A^T{\left(A{A}^T\right)}^{-1}$$ (8)

where $x_p$ indicates the best producer's position; $x_{worst}$ denotes the global worst position; A denotes a matrix of $1\times d$. The initial position is generated as followings (Li et al., 2022e):

$$x_{i,j}^{t+1}=\{\begin{array}{c}\begin{array}{cc}{x}_{best}^t+\lambda \left|x_{i,j}^t-x_{best}^t\right|,& f_i>f_g\end{array}\\ \begin{array}{cc}{x}_{i,j}^t+k\frac{\left|x_{i,j}^t-x_{worst}^t\right|}{f_i-f_w+\epsilon },& f_i=f_g\end{array}\end{array}$$ (9)

where λ is the step control function; $f_i$ denotes the current value; $f_g$ or$f_w$ represents the global best or worst adaptation value; k denotes the direction of sparrow movement; $x_{best}$ denotes the global optimal position; ε is the minimum value, avoiding the divisor to be zero.

3.4. Elman neural network (ENN)

ENN is a recursive network featured by an internal self-referencing layer. ENN consists of four components: input layer, context layer, hidden layer, and output layer (Boualem et al., 2022). The context layer is designed to store or memorize the output values before the hidden layer. The outcome of the hidden layer is transferred to the input layer through the context layer. This feedback mechanism makes it possible to process more dynamic messages, realizing the dynamic operation, shown in Fig. 1 .

Fig. 1.

Structure of Elman neural network.

The nonlinear expression of the ENN is as follows (Boualem et al., 2022; Ruiz et al., 2018)：

$$y\left(k\right)=g\left(w^{3}x\left(k\right)\right)$$ (10)

$$x\left(k\right)=f\left(w^1{x}_c\left(k\right)+w^{2}u\left(k-1\right)\right)$$ (11)

$$x_c\left(k\right)=x\left(k-1\right)$$ (12)

where k is the training number of ENN; y denotes the n-dimensional output vector; x represents the hidden layer neuron output vector; x_c denotes the feedback state vector; u means the input vector; g represents the transfer function of the output neuron; $w^1$、 $w^2$ and$w^3$ is the connection weight matrices, and f represents the transfer function of the hidden layer neuron.

3.5. GRA-ENN-SSA hybrid model

In this work, the proposed model is shown in Fig. 2 . GRA is introduced to obtain data on indicators related to carbon emissions. To address the defect that the ENN model is prone to drop to local optimum resolution, the SSA is designed to optimize ENN to enhance the model's accuracy.

Fig. 2.

Process of GRA-ENN-SSA algorithm.

The model was completed by MATLAB 2021b software. The evaluation of model performance takes mean absolute percent error (MAPE), mean absolute error (MAE), and root mean squared error (RMSE) as indicators to assess the deviation between the actual value and the predicted outcome. The smaller the value of the indicators above, the better the regression ability and stability of the model GRA-ENN-SSA. The formulas of the indicators are expressed as:

$$RMSE=\sqrt{\frac{1}{T}}\sum _{i=1}^T{\left(y\left(i\right)-\hat{y}\left(i\right)\right)}^2$$ (13)

$$MAPE=\frac{1}{T}\sum _{i=1}^T\left|\frac{y\left(i\right)-\hat{y}\left(i\right)}{y\left(i\right)}\right|\times 100\%$$ (14)

$$MAE=\frac{1}{T}\sum _{i=1}^T\left|y\left(i\right)-\hat{y}\left(i\right)\right|$$ (15)

where $y\left(i\right)$ is the actual value and$\hat{y}\left(i\right)$ is the predicted value.

3.6. Data source

The data employed in the work are from 1995 to 2020, where GDP, PGDP, SSP, TSP, FAI, TP, UR, OL, and RI are taken from the China Statistical Yearbook and National Bureau of Statistics; TEC, CCP, NFC, EI, FFP, TPP, and EC are collected from the China Energy Statistical Yearbook, and the data of carbon emissions (CE) are gained from statistics of Organization for Economic Cooperation and Development (OECD) and the BP Energy Statistical Yearbook.

4. Results and discussion

4.1. Filtering of carbon emission influencing factors

Carbon emissions are subject to multiple elements in multiple systems such as economics, society, and energy. Bringing all the factors in Table 1 into the model directly would certainly make the model even more complicated and hard to run. To eliminate irrelevant criteria and enhance operational efficiency, indicators related to carbon emissions were quantified by the GRA model. The distributions of the data are given in Fig. 3 . The correlation of indicators is shown in Fig. 4 .

Fig. 3.

Distribution of factors.

Fig. 4.

Correlation degree of factors.

Factors with a correlation degree of more than 0.7 are identified as input indicators for the prediction model. After the screening process, the GDP per capita, population, urbanization rate, total energy consumption, the proportion of the tertiary industry, the share of non-fossil energy, total electricity consumption, and R&D inputs were confirmed as input parameters. Looking at the history of the above factors (Fig. 3)， China's total population, per capita GDP and urbanization rate gradually increased from 1995 to 2020. Total energy consumption, total electricity consumption and the proportion of non-fossil energy all showed an upward trend. Although the structure of the secondary industry has some fluctuations, the overall trend is downward. The above factors were selected as the main driving force of carbon emissions, which also reflected that China relied on the rough economic development mode in the past to consume huge energy (especially fossil energy) and bring huge carbon emissions.

4.2. Performance assessment

To test the performance of the proposed model, the historical data from 1990 to 2015 were regarded as the training set, and the others (2015–2020) were treated as the test set. By inputting normalized data (see Fig. 3), China's carbon emissions from 2016 to 2020 were forecasted by employing ENN, GRA-ENN, GRA-BP, GRA-SSA-BP, and GRA-SSA-ENN. Among them, the BP transfer function uses the Sigmoid function. And the neurons number of in the output layer is 1. In the input layer, the neurons' number of ENN and SSA-ENN, GRA-SSA-ENN is equal to the number of influencing factors. In the work, it is set as 8. The number of training and accuracy requirements are kept consistent, the training times are 200. The above model is run 50 times. The average value of 50 operation results is treated as the outcome.

The outcomes of five models for TCE are presented in Fig. 5, Fig. 6 . The MAE, RMSE, and MAPE of GRA-ENN show smaller than those of the ENN. It reveals that the screening of indicators by GRA improves the accuracy of model prediction; The comparison shows that the prediction error of GRA-SSA-ENN for carbon emissions is the smallest among the five algorithms. It indicates the excellent accuracy and superiority of the model proposed in this work, and by optimizing the parameter settings of the model, it can achieve superior prediction performance and can be employed to research carbon emissions.

Fig. 5.

Comparison of simulation results.

Fig. 6.

Comparison of prediction errors.

4.3. Scenario designing

According to the Chinese government's development plan and policy measures at each stage, while referring to the forecast results of authoritative institutions at home and abroad, the influencing factors are distinguished into three growth modes: high, medium, and low. We explain the variable setting in the medium scenario. The high and low scenarios are developed from the middle scenarios and correspond to the characteristics of China's future social-economic development. The parameters of scenarios are described in the Supplementary data Tables 1–8

• 1.Population (Niu et al., 2020; Xu et al., 2019): As stated in China's Population Development Plan 2016–2030, the national population is expected to be 1.42 billion in 2020, and 1.45 billion in 2030. After that, the population continues to drop to 1.37 billion by 2050. The annual growth rate of the Chinese population shows around 0.7% from 1990 to 2019. In this study, we regard the population growth rate at 0.5% in the medium scenario for 2021–2030, followed by an average growth rate of −0.5%, −0.4%, and −0.3% for the following decades to 2060. The low and high scenario parameters fluctuate above and below 0.1%.
• 2.GDP per capita (Xu et al., 2019): Based on data from the past decades, a GDP per capita growth rate of 5.5% is set for the medium scenario for 2021–2030. Given that economic growth and industrialization rates decline after rapid growth (Xu et al., 2019), a decline of 1% per decade is set afterward. Given the impact of the COVID-19 epidemic, the parameters of the high and low are floated up by 0.3%, and 0.5%.
• 3.Urbanization rate (Yuan et al., 2014): In line with National Population Development Plan (2016–2030), China's urbanization rate could hit 70% by 2030, and 80% by 2050. This study assumes that the urbanization rate shall remain stable and grow moderately after 2050. The target calls for the growth of urbanization to be 1.0% from 2021 to 2025, 0.8% from 2030 to 2050, and then decrease to 0.6%. The low and high scenario parameters fluctuate above and below 0.2%, respectively.
• 4.Total energy consumption (Niu et al., 2020): Taking into account the projections of the World and China Energy Outlook 2050, this study supposes that China's energy consumption will maintain a positive growth at a rate of 1.4% in the medium scenario from 2020 to 2030, and at −0.04% and −0.5% in the next two decades, respectively. The rest of the scenery moves up and down by 0.4%.
• 5.Secondary industry structure (Fang et al., 2019): Considering the economy and society hit a threshold of urbanization and industrialization, the secondary industry growth will slow down, whilst the growth of the service industry will raise (Xu et al., 2019). Thus, following the assumptions of (Yuan et al., 2014) and the impact of the COVID-19 epidemic, we argue that the growth of the secondary industry in the medium case is −2.0%, −1.5%, and −1.0% for 2021–2030, 2031–2040, and 2041–2060, respectively. The low and high scenario parameters fluctuate above and below 0.5%.
• 6.Proportion of non-fossil energy: Given China Energy and Power Development Outlook, the work pointed out that the share of non-fossil energy consumption in energy usage in China could hit about 22%, 40%, 69%, and 81% in 2025, 2035, 2050, and 2060, respectively. For this study, we take these parameters as the medium scenario.
• 7.Total electricity consumption: The Chinese government's 14th Five-Year Plan (FYP) states nation's electricity consumption will be between 9 trillion and 10 trillion kWh by 2025, with a growth rate of 4%–6%. Also, China Energy and Power Development Outlook point out that China's electricity demand will continue to grow, with a gradual slowdown, and enter a saturation growth stage after 2035; total electricity consumption will be around 9.8 trillion kWh, 12.4 trillion kWh, 13.9 trillion kWh and 13.3 trillion kWh in 2025, 2035, 2050 and 2060, respectively. The parameters mentioned above are considered medium scenario parameters. The low and high scenario parameters fluctuate above and below 1 trillion kWh, respectively.
• 8.Science and technology level (Li et al., 2022a): Investment in research and experimental development is an important source of science and technology funding. In 2020, China's R&D investment intensity exceeded 2.4%. It is still far from the level of 2.5%–4% in developed countries. The R&D intensity will be 2.8% in 2030. The annual growth regarding R&D intensity under the medium scenario is 2.3% in 2021–2030, 2.5%, 2.2%, and 2.0% in 2030–2040, 2040–2050, and 2050–2060, respectively. The low and high scenario parameters fluctuate above and below 0.5% and 0.2%.

Given the CCUS technologies, a low development scenario and a high development scenario are created based on future CCUS in China in related research (Sun et al., 2022b), and the parameters are set in Table 2 .

Table 2.

Parameters of CCS in 2021–2060 (unit: %).

Time	2021–2025	2025–2030	2030–2040	2040–2050	2050–2060
High	7.5	12.7	17.5	22.5	22
Low	5	10	15	20	17.5

Given the settings in existing work (Li et al., 2022a; Ren and Long, 2021; Xu et al., 2019), to reflect the potential carbon-neutral path, nine scenarios are developed based on the carbon reduction policy implementation, economic development pattern, and social conditions. The parameters of the scenarios are set in Table 3 . S1 is the slowdown scenario, the economy is in recession, with all indicators at their lowest; S2 is the general scenario, and as usual, the emphasis is purely on economic development; S3 is the green development scenario, focusing on the quality of economic development and emphasizing low-carbon production; S4 is the energy saving scenario, slowing down large-scale development and construction, increasing energy consumption costs; S5 is the energy economy transition scenario, highlighting the upgrading of industrial structure and the use of new energy sources; S6 is the weak emission reduction scenario, secondary industry dominates, with slow development; S7 is the strong emission reduction scenario, emphasis on emission reductions while restraining economic development; S8 is the medium emission reduction scenario, industrial structure upgrade but the urbanization process slowly. and S9 is the intensive industry scenario, rapid urbanization, but slow industrial upgrading.

Table 3.

Parameter designing of scenarios.

Scenario	POP-	PGDP	UR	TEC	TSP	NFP	EC	RI	CCUS
S1	Low	Low	Low	Low	Low	Low	Low	Low	Low
S2	Medium	Medium	Medium	Medium	Medium	Medium	Medium	Medium	Low
S3	High	High	High	High	High	High	High	High	High
S4	Low	Low	Low	Medium	High	Medium	Medium	Medium	High
S5	Medium	High	Medium	High	High	High	High	Medium	High
S6	Medium	Low	Medium	Low	Low	Low	High	Low	Low
S7	High	Low	High	Medium	Medium	High	High	High	High
S8	Low	Medium	Low	High	Medium	Medium	Medium	Medium	Low
S9	High	High	Medium	High	Low	Low	Low	Medium	High

4.4. Analysis of carbon neutral scenarios in China

The proposed model has been deployed to simulate the evolution and scenarios of carbon emissions in China from 2020 to 2060. Fulfillment of emission reduction aims is presented in Table 4 and Fig. 7 . Obviously, the timing of the peak carbon emissions under nine scenarios varies from 2029 to 2034, and the peak values range from 11568.6 to 132224.2 Mts. Three categories of the nine scenarios can be distinguished concerning the completion of the carbon peak and neutrality targets.

Table 4.

Prediction results in different scenarios.

Scenario	Peak time	Peak value/Mts	Neutrality in 2060	CE in 2060/Mts
S1	2034	12998.4	NO	6694.7
S2	2032	12553.5	NO	3327.4
S3	2028	11918.3	YES	−23.8
S4	2030	11568.6	NO	1495.7
S5	2029	12330.5	NO	599.3
S6	2032	12493.5	NO	6380.8
S7	2030	12151.2	NO	2167.9
S8	2032	12776.6	NO	4364.2
S9	2032	13224.2	NO	5081.3

Fig. 7.

China's CO2 emissions path during 2020–2060 under different development scenarios.

The first group includes S1, S2, S6, S8, and S9. In this category, carbon emissions are promising to get peak after 2030, and carbon neutrality will not occur in 2060. Here, there is rapid economic and social development, sluggish urbanization, and insufficient industrial and energy restructuring. Neither peak nor net-zero emissions can be fulfilled as expected. The second category covers S4, S5, and S7. In this group, all scenarios allow for a peak in 2030 or earlier, but there is still a gap in attaining carbon neutrality. Distinct from the first category, the government will impose carbon taxes and other mandatory measures to implement an energy conservation strategy, facilitate the transformation of the energy structure, and optimize the foreign trade structure. Thus, compared to the former, the time to peak and the amount of carbon emissions in this category are both better than the former; the value of emissions in 2060 is also smaller. The third type includes S3, and only S3 succeeds in getting carbon neutrality in 2060. Compared with the above categories, the government in this scenario attaches more value to the pace and level of economic development. With the application of effective emission reduction measures and CCUS, the targets can be fulfilled as scheduled.

Regarding the development stages, China's carbon mitigation presents distinct features in the nine scenarios from 2020 to 2060, as shown in Fig. 8 . To begin with, China's carbon emissions take the first step into the plateau in the S3 and S5, and the remaining scenarios reach this position as late as the 16th FYP. Thereinto, S1, and S7 present a downward shift towards carbon emissions after the 16th FYP, lagging behind all the other scenarios. The emissions will drop after the peak, and the decline will reach its maximum during the 19th FYP, with S3 being the most noticeable. S1 has the least amount of emission reduction. As a consequence, the cumulative reduction by 2060 appears sharply diverse among the nine scenarios. Specifically, the cumulative CO₂ emission reduction is 2913 million tons in the S1 and 9855 million tons in the S3, while it ranges from 3495 to 9143 million tons for the remaining scenarios.

Fig. 8.

Emission reductions of each scenario during the 14th-21st FYP.

4.5. Discussion

China has a great potential to meet peak carbon emissions by 2030, which shows no difference from the results of the literature (Fang et al., 2019; Niu et al., 2020). The earliest peak time and peak value we predict is in 2028, and our results are distributed in the range of previous studies' estimations. Many scholars have estimated that optimistically China will peak carbon emissions in 2025 (Zhou et al., 2019), or 2029 (Niu et al., 2020). The latest time to peak is around 2048 (Hao et al., 2022). In addition, the peak value is also distributed in the range of 10 billion (Wu and Xu, 2022) to 19 billion tons (Hao et al., 2022).

Yet dramatic efforts need to be devoted to realize carbon neutrality by 2060. For the neutral scenario, the energy consumption intensity in China would decline by about 30% in 2030, 65% in 2050, and 80% in 2060; the percentage of non-fossil fuels consumption would rise to 25% in 2030, and grow to more than 65% in 2050 and 80% in 2060; and energy consumption should be controlled at 6.5 billion tons coal, which is a great challenge for the existing industrial and energy structure of China. In addition, total energy consumption is most closely related to carbon emissions, which also shows that China's fossil energy-based energy structure has not been substantially changed and is still the main driver of carbon emissions. Therefore, total energy consumption should be controlled. However, China's energy consumption is still increasing, and the supply of clean energy is sluggish, which is a tough challenge for China to mitigate carbon emissions. At the same time, the urbanization process has significantly contributed to carbon emission reduction, and also indicates that China has embarked on improving energy use efficiency after reaching a certain level of urbanization.

5. Conclusions and policy recommendations

5.1. Conclusions

Given the raging COVID-19 epidemic, the economic development pace and production patterns of China have been hit gravely, and there are great uncertainties about whether China can realize carbon neutrality by 2060. To this end, this study develops a hybrid model integrated with GRA, SSA, and ENN to investigate the potential of China to achieve carbon neutrality and some conclusions are drawn: (1) Through the evaluation results of 16 factors related to carbon emissions by GRA, it reveals that GDP per capita, population, urbanization rate, total energy consumption, the share of tertiary industry, the share of clean energy, total electricity consumption, and technology level are highly correlated with carbon emissions; (2) Results of GRA-ENN and ENN prediction suggest that MAPE, MAE and RMSE of the former are better than those of the latter. It reveals that the screening of indicators by GRA improves the performance of model prediction; the prediction accuracy of the GRA-SSA-ENN model is remarkably superior to the other four models. That is, the proposed model shows excellent performance in carbon emission prediction; (3) Simulation results of multiple scenarios reveal that China has a promising prospect of reaching a peak from 2028 to 2030, with a value of about 11,568.6 to 12,330.5 Mt. Moreover, only the green development scenario allows for carbon neutrality as scheduled. By the time 2060, the share of renewable energy on the energy supply side will exceed 80% and the energy consumption intensity could decline by 80%.

5.2. Policy recommendations

In this regard, this work puts forward several recommendations: (1) To enhance energy utilization efficiency and promote a continuous reduction of energy consumption via industrial upgrading and structural adjustment and technological energy-saving measures. In addition, China should optimize the energy structure with renewable energy as the primary power source, diversify energy utilization, and propel the renewable energy consumption share in primary energy. (2) Given the latecomer status of CCUS implementation in China, policies to support the progress of CCUS technology are needed to achieve rapid application in industrial sectors, such as power and steel. (3) Ramp up the pace of carbon trading market construction and carbon sink subsidy mechanism, allowing the market to exert its role in allocating resources. It will restrain emissions and promote carbon mitigation with the market mechanism.

In brief, the proposed model enriches the methodology in the field of carbon emission prediction and provides a valuable reference and tool for exploring the path of carbon neutralization. In the future, we will further discuss the economic and environmental costs to achieve carbon neutrality. In addition, we will also conduct an in-depth discussion on the path to carbon Neutral in each province of China and the socio-economic conditions that need to be met.

Credit author statement

Yisheng Liu: Conceptualization; Methodology; Funding acquisition; Supervision. Meng Yang: Data curation; Writing - original draft; Formal analysis; Software; Writing - review &editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix ASupplementary data

The following is the Supplementary data to this article:

Data availability

Data will be made available on request.