Development Level Evaluation and Driving Factors Analysis of China’s New Energy System: Based on Random Forest

Abstract

Sustainable utilization of energy depends on the establishment of an advanced energy system. As the world’s largest consumer and importer of energy, China’s progress in this field has attracted considerable attention. This study seeks to address the limitations of most existing research, which largely remains at a qualitative level, by expanding perspectives and methodologies. Utilizing think-tank research approaches and indicator system evaluation methods, it quantitatively evaluates the development level of new energy systems across thirty provincial-level administrative regions in China from 2011 to 2023. Machine learning methods were applied to empirically analyze the driving mechanisms of “new” factors through the construction of a random forest model. The results reveal that: (1) China’s new energy system exhibited an overall positive development trend, albeit at a relatively slow pace and with notable spatial disparities. The development levels of the three core objectives followed a gradient pattern, showing marked improvements after the implementation of China’s supply-side structural reform policies. (2) Innovation funding and high-level labor input served as the dominant driving forces for development, while factors such as the scale of the technology market, the proportion of the tertiary sector, and environmental regulation investment played supplementary roles, with regional variations observed.

1. Introduction

Since the signing of the Paris Agreement in 2015 and the outbreak of the Russia–Ukraine war in 2022, the global energy landscape has undergone fundamental disruption. The inherent weaknesses of the global energy system—including mismatched supply and demand, policy conflicts, and developmental disparities—have become increasingly evident [1,2]. Energy competition has shifted from traditional geopolitical struggles toward the transformation of national energy systems [3].

China ranks first globally in both energy consumption and imports. “Accelerating the planning and construction of a new energy system” was highlighted as a strategic direction in General Secretary Xi Jinping’s report to the 20th National People’s Congress (NPC) of China. This underscores that China is at a crucial stage in transitioning from a traditional to a new energy system. The path of this reform will profoundly influence the global energy landscape.

At present, academia has not reached a precise definition of the “new energy system.” Internationally, the development direction varies across nations due to differences in energy endowments [4], geopolitical contexts [5], technological capabilities [6], and socio-economic foundations [7]. For instance, the United States, pursuing “energy independence” as a strategic goal, utilized new extraction technologies to launch the shale oil and gas revolution, becoming the world’s largest oil and gas producer and significantly strengthening its global energy influence [8]. In contrast, the UK has advanced policies tied to its “net-zero emissions target,” strongly promoting wind, nuclear, hydrogen, and marine-integrated energy technologies to achieve a super-clean energy system [9,10].

China’s energy system reform faces more severe challenges, reflected in several aspects. First, its energy endowment is unbalanced, with abundant coal resources but limited oil and natural gas reserves. Heavy reliance on imported oil and gas creates substantial risks for energy security [11,12]. Second, as the world’s largest developing nation and carbon emitter, China confronts dual pressures: the demands of economic growth and the obligations of international carbon reduction commitments, resulting in significant internal contradictions [13,14,15,16]. Third, future economic development will revolve around “New Quality Productivity,” a concept of modern productivity emphasizing advanced technology, efficiency, and quality [17,18], while prioritizing green and sustainable practices [19,20]. This places higher requirements on energy utilization models, conversion efficiency, and the management of externalities.

Given China’s central role in global carbon reduction efforts, the practical energy needs of developing countries, and the constraints imposed by its geography, demography, technological base, and resource endowment, research on China’s new energy system carries far-reaching implications. It offers insights for global energy system evolution, contributes to mitigating supply risks, and ultimately supports the achievement of sustainable development goals for humanity.

Current academic research on China’s new energy system can be broadly categorized into two aspects. First, scholars have defined the essence of China’s new energy system based on the content of major annual conferences and reports, and through analyses of internal and external environments as well as historical data, and have subsequently proposed recommendations for future development strategies. For example, Guilhot argued that China must address institutional, economic, and environmental challenges to construct a sustainable energy system [15]. Li et al. developed a framework for China’s new energy system focused on energy security, integrating both internal and external dimensions by applying think tank technology [21]. Zhu et al. examined historical trends in China’s energy consumption, emphasizing that the development of a new energy system should prioritize technological innovation and the achievement of “dual carbon goals” [22]. Fan et al. further highlighted the importance of intelligent systems for advancing China’s energy innovation [23].

Second, some scholars have explored the intersection between China’s energy structure transformation and other domains, thereby identifying the developmental direction of China’s new energy system. For instance, Zou et al. emphasized that establishing a new energy system is crucial for realizing China’s new quality productivity [24], and proposed the “energy triangle theory,” which suggests that the system should be constructed around the principles of greenness, economy, and security [25]. Zhang et al. [26] and Fang et al. [27] presented similar perspectives and supplemented them with empirical evidence. Li et al. demonstrated that energy structure transformation contributes to enhancing urban resilience, arguing that China’s energy transition should prioritize green development and resilience [28]. Wang et al. also expressed comparable views using different methodological approaches [29].

In summary, research on China’s new energy system has primarily concentrated on green transition, energy security, developmental contradictions, intelligent energy, and its role in advancing new productive forces and urban development. However, the main shortcomings can be identified as follows: (1) studies have largely remained at a theoretical level, lacking quantitative evaluation, thereby limiting empirical exploration; (2) scholars have focused on the effects of energy structure transformation on other sectors, yet systematic examination of the internal mechanisms driving the new energy system itself remains insufficient. This gap has hindered the empirical foundation necessary for shaping future development pathways.

To overcome these limitations, the present study applies think tank technology and the Analytic Hierarchy Process (AHP) to construct an indicator evaluation system for assessing the development level of China’s new energy system. The entropy weight method was used to measure the development levels across 30 provincial-level administrative regions from 2011 to 2023. Furthermore, a random forest model was employed to analyze the driving mechanisms and regional disparities underlying this development.

The marginal contributions of this article are twofold: (1) starting from the current development objectives of China’s energy system, it constructs an indicator evaluation system for assessing development levels based on think tank technology and the “impossible triangle” theory. This approach addresses the lack of quantitative evaluation in existing research and provides new insights into building indicator evaluation systems in this field; (2) by employing machine learning techniques, this study supplies empirical support for policy research on China’s new energy system, identifying key drivers that embody its “new” characteristics. This avoids the constraints of traditional linear regression models and introduces novel perspectives for related areas of study.

2. Research Area and Data Information

China currently consists of 34 provincial-level administrative regions. Due to substantial gaps in statistical data for Xizang and limited accessibility of data for Hong Kong, Macao, and Taiwan, these four regions were excluded from the study sample. The remaining 30 regions were selected as the research areas, with their specific geographic boundaries illustrated in Figure 1.

The data used in this article are primarily sourced from the China Energy Statistical Yearbook, China Statistical Yearbook, China Labour Statistical Yearbook, China Statistical Yearbook on Environment, China Statistical Yearbook on Science and Technology, provincial statistical yearbooks, and the Annual Development Report of China’s Electric Power Industry. To address missing or inaccessible data, appropriate imputation techniques were employed according to the context and availability of information. The mean value method and interpolation were used to estimate missing data points in line with observed temporal patterns and actual regional conditions. Specifically, for indicators with less than 5% missing data in a given year and an approximately normal distribution, we take the mean of all available data to fill in the gaps. An example is the environmental regulation investment variable across regions in 2011. We determined for missing values in specific years for a given region via a two-point linear equation on the basis of data from the preceding and following years. For example, the missing data for the total length of transmission lines indicator in Yunnan Province for 2015 can be interpolated linearly via data from 2014 and 2016. In addition, outliers detected for specific years were adjusted to align with historical trends, ensuring continuity and reliability of the dataset. To enable standardized processing using the entropy weight method, all values less than or equal to zero were corrected through non-negative translation techniques.

3. Method

3.1. Development Level Evaluation of China’s New Energy System

3.1.1. Construction of the Indicator Evaluation System

To ensure the relative independence of each element and avoid omitting key variables, the principles of the Analytic Hierarchy Process (AHP) were applied to construct an indicator evaluation system for assessing the development level of China’s new energy system. The specific steps were as follows: first, identify the target layer; second, decompose it into a criterion layer based on the essence of each target; and finally, select appropriate measurement indicators for each criterion.

Focusing on the core development requirements of China’s new energy system, as defined by both policy and academic discourse, text-mining techniques in think tank research were employed to identify 52 Chinese and English core journal articles since 2023 that describe China’s development goals for its new energy system. Similar viewpoints were integrated and ranked by frequency of occurrence. Among these, “green,” “security,” “economy,” “sustainability,” and “intelligence” emerged as the top five. The distribution of literature quantity is presented in Figure 2.

As shown in Figure 2, “green” is regarded as the primary requirement of China’s new energy system, with its core emphasis on low-carbon and low-pollution development [22,30,31,32,33]. The second-highest focus is “sustainability.” However, this concept is overly broad and often overlaps with other objectives, typically serving as a general summarizing principle in the literature; hence, it is unsuitable as a sub-objective in the indicator system. In contrast, “security” and “economy” have clearer and more distinct connotations, making them more appropriate for the objective layer. “Security” emphasizes the long-term, stable supply capacity of the energy system [34,35], while “economy” relates to reducing production costs and supplying energy at lower prices [25]. The concept of “intelligence” is less frequently highlighted and is generally described as supporting “security” [23,36]. Therefore, “intelligence” was considered more suitable as a criterion within the “security” dimension.

Based on these considerations, the development objectives of China’s new energy system are proposed to be divided into three dimensions: green, security, and economy. This perspective corresponds to the three elements of the “impossible triangle theory,” which states that policies related to societal development cannot simultaneously meet the goals of greenness, economy, and security [37]. By contrast, the new energy system demands that all three objectives be satisfied concurrently. This perspective aligns with the “energy triangle theory” proposed by Zou et al. [25] and serves as the basis for the transformation from a traditional to a new energy system.

Further analysis was carried out to determine the criterion layer of these three objectives. For Green, the system emphasizes low-carbon and low-pollution practices in both production and consumption. The zero-carbon features of new energy and the development of advanced pollution-prevention technologies are critical tools for achieving this goal [32]. For economy, industries such as wind and solar energy have already achieved economies of scale in many regions of China, enabling power supply at costs lower than those of traditional energy sources [25]. In addition, improvements in energy utilization efficiency [38] and energy conservation [39] remain crucial. For security, an integrated network where sources, transmission, loads, and reserves are mutually coordinated is essential [21,40]. At the same time, the application of digital technologies with high computational power, precision, and interference resistance enhances system capabilities in information transmission, emergency response, and comprehensive control, thereby raising the overall security level [36].

In summary, a schematic diagram of China’s energy system transformation is shown in Figure 3, while the indicator evaluation system for assessing the development level of China’s new energy system is provided in

Table 1. Indicator evaluation system for the development level of China’s new energy system.

Target Layer	Criterion Layer	Indicator Layer 1	Indicator Layer 2	Nature	Weight	Data Source
Green	Low-carbon	Level of carbon emissions reduction	CO2 emissions/GDP	Negative	4.17%	②③
	Low-pollution	Industrial exhaust emission levels	SO2 emissions/GDP	Negative	2.54%	②③
		Industrial effluent discharge levels	Industrial effluent discharge volume/GDP	Negative	3.36%	②③
	Pollution prevention capabilities	Industrial exhaust treatment capability	Number of industrial exhaust treatment facilities	Positive	8.03%	②
		Industrial effluent treatment capacity	Number of industrial effluent treatment facilities	Positive	9.51%	②
Economy	Utilization of New Energy Sources	Share of new energy consumption	Wind and solar power generation/Total electricity consumption	Positive	13.18%	③
	Energy Utilization Efficiency	Energy consumption intensity	Total Energy Consumption/GDP	Negative	6.39%	①③
		Energy conversion Efficiency	Energy Processing and Conversion Output/Input	Positive	5.34%	①⑥
	Energy Conservation	Energy conservation level	Reduction rate of energy consumption of 10,000 Yuan GDP (equivalent value)	Negative	3.30%	⑥
Security	Energy Network	Scale of electric transmission lines	Total Length of transmission lines/Regional area	Positive	9.72%	⑦
		Scale of highways and railways	Total length of graded highways and railways/Regional area	Positive	3.68%	⑥
	Energy load	Final electricity consumption	Per capita electricity consumption	Negative	3.08%	⑥
	Intelligence	Enterprise Digitalization Level	Enterprise Digitalization Index	Positive	12.14%	⑤⑥
		Robot Installation Density	Number of robots/Number of employees	Positive	7.15%	④
	Energy reserves	Electricity reserve capacity	Installed power generation capacity	Positive	3.35%	⑦
	External dependence	Share of imported energy consumption	Energy imports/Total regional energy consumption	Negative	5.06%	①

Note: All indicators involving energy quantities are measured in ten thousand tons of standard coal. The calculation method of indicator weights is described in Section 3.1.2. GDP refers to gross domestic product. Data sources: ① China Energy Statistical Yearbook; ② China Statistical Yearbook on Environment; ③ China Statistical Yearbook; ④ China Labor Statistical Yearbook; ⑤ China Statistical Yearbook on Science and Technology; ⑥ Provincial statistical yearbooks; ⑦ Annual Development Report of China’s Electric Power Industry.

. The calculation method for the digitalization index follows the approach of Wu et al. [41].

3.1.2. Comprehensive Evaluation Based on the Entropy Weight Method

To minimize the influence of subjective factors on evaluation results, and following the approach of Li et al. [28], the entropy weight method was adopted to comprehensively evaluate the development levels of new energy systems across China’s 30 provincial-level administrative regions. The specific steps are as follows [42,43]:

Step 1: All the data are standardized using the extreme variance method to ensure consistency in the measurement of all the indicators, and X_ij denotes the value of indicator j of region i, X_ij’ represents the standardized value of X_ij. For convenience, X_ij’ will be treated as X_ij.

Step 2: For target layer, there are m samples that are evaluated, and each sample corresponds to n evaluation indicators, constructing a m × n data matrix (X_ij) m × n.

Step 3: Calculate the information entropy e_j for indicator j:

$$e_j=-{\displaystyle \raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$\ln m$}\right.}{\displaystyle {\sum }_{i=1}^m{P}_{ij}\ln P_{ij}}$$

(1)

where P_ij is the weight of indicator j of region i, which is calculated as:

$$P_{ij}=X_{ij}/{\displaystyle {\sum }_{i=1}^m{X}_{ij}}$$

(2)

The entropy value e_j indicates the degree of dispersion on the data of the jth indicator. Indicators with greater variation or higher order are considered to possess greater evaluative capability.

Step 4: Calculate the weights w_j for indicator j:

$$w_j=(1-e_j)/{\displaystyle {\sum }_{j=1}^n(1-e_j)}$$

(3)

Step 5: Calculate the combined evaluation value NES_i that indicates the development level of the new energy system for the sample m:

$$NE{S}_i={\displaystyle {\sum }_{j=1}^n{w}_j{X}_{ij}}$$

(4)

Based on the entropy weight calculations, the weight distribution diagram for the three target dimensions of China’s new energy system was generated, as shown in Figure 4. The results indicate that the weights of the green and economy targets are relatively close, while the security target carries a significantly higher weight. This suggests that security plays a more decisive role in the development of China’s new energy system.

3.2. Random Forest Model

Random Forest is a machine learning model built upon feature randomization and the bagging algorithm. Its core principle is to construct multiple decision trees capable of learning and to average their prediction results, thereby achieving an ensemble evaluation superior to that of a single decision tree in terms of both robustness and accuracy [44,45]. Compared with traditional linear regression models, random forests reduce sensitivity to outliers and avoid problems of multicollinearity among variables, making them well-suited for analyzing the driving mechanisms of composite dependent variables [46]. Accordingly, this study employs a random forest model to analyze the drivers underlying the development level of China’s new energy system.

3.2.1. Identification of Driving Factors

The driving factors of an energy system should be closely aligned with the intrinsic properties of energy [21]. Based on these attributes, the following driving factors were selected for the development of the new energy system:

Resource and economic attributes: Energy serves as a fundamental input for industrial production, reinforcing its economic nature [47]. According to neoclassical growth theory, economic growth is determined by capital and labor. Similarly, the development of energy systems requires both capital and labor inputs. Consequently, innovation fund investment and high-level labor input were identified as key drivers of the transition from traditional to new energy systems, with technological innovation at the core [48,49,50].

Commodity attributes: As a critical traded commodity, energy system development is influenced by market size. Market expansion facilitates integration across regulatory bodies, industries, and enterprises, promoting systemwide coordination [21,51]. To highlight the distinct characteristics of the new energy system, the technology market size was selected in place of general market size.

Industrial attributes: The capacity of the energy industry significantly impacts regional industrialization [52]. Traditional energy development has relied heavily on secondary industries such as mining, chemicals, manufacturing, and energy supply. In contrast, the development of new energy systems places greater emphasis on the tertiary sector, including optimizing resource allocation in energy markets through finance [53], raising public awareness of energy conservation and environmental protection through education [54], and accelerating R&D via scientific research and technical services [55]. Hence, the share of the tertiary sector was incorporated as a driving factor.

Finally, to better reflect the stronger emphasis on ecological and environmental considerations in the new energy system compared with the traditional one, environmental regulation investment was added as an additional driving factor. In summary, the driving factors and their measurement indicators for both traditional and new energy systems are presented in

Table 2. Driving factors and measurement indicators for the development of traditional and new energy system.

Traditional Energy System	New Energy System
Driving Factors	New Driving Factors	Indicators	Abbreviation
Fund input	Innovation fund input	Total innovation funding of industrial enterprises above designated size	FUN
Labor input	High-level labor input	Full-time equivalent R&D personnel in industrial enterprises above designated size	HUM
Market size	Technology market size	Total value of technology market transactions	MAR
Share of secondary industry	Share of tertiary industry	Share of employment in the tertiary sector	IND
Without considering environmental issues	Environmental regulation investment	Share of environmental governance investment in government expenditure	ENV

.

3.2.2. Construction of the Model

Following the previous scholars [44,45,56], the construction of the random forest model was carried out in the following steps:

Step 1: Establish the training dataset N_T for each driving factor variable and the target prediction variable under a given time series:

$$N_T=\left\{V_T(x_i,y),i\in 1,2,\cdot \cdot \cdot ,5\right\}$$

(5)

where V_T(x_i, y) denotes all combinations of driving factor variables and target prediction variables under a given time series T; x_i denotes the five driving factor variables selected in this article (FUN, HUM, MAR, IND, ENV); and y reflects the evaluation results for development level of China’s new energy system.

Step 2: Generate sub-datasets using bootstrap sampling. The purity of split nodes was measured by the Gini index:

$$Gini(P)=1-{\displaystyle {\sum }_{k=1}^K{P}_k^2}$$

(6)

where P_k denotes the proportion of samples belonging to category k in the node; K is the number of categories.

Step 3: Employ the Gini coefficient as the dependency criterion for generating branches in decision trees. The aim is to select appropriate split points and driving factors X_j to ensure that the Gini coefficient is minimized. If the dataset N of the jth driver variable X_j is split into N_left and N_right, then the Gini index after splitting is:

$$Gin{i}^{\prime }={\displaystyle \frac{\left|N_{left}\right|}{\left|N\right|}}Gini(N_{left})+{\displaystyle \frac{\left|N_{right}\right|}{\left|N\right|}}Gini(N_{right})$$

(7)

During this process, variance must be minimized through splitting. The formula for calculating the mean square error (MSE) is:

$$MSE={\displaystyle \frac{1}{\left|N\right|}}\left[{\displaystyle \sum _{x_i\in N_{left}}{\left(y_i-{\overline{y}}_{left}\right)}^2+{\displaystyle \sum _{x_i\in N_{right}}{\left(y_i-{\overline{y}}_{right}\right)}^2}}\right]$$

(8)

where y_left and y_right denote the two sub-datasets.

Step 4: Calculate the mean of the results from the n decision trees generated by splitting to produce the final model’s evaluation value, using the following formula:

$$Y_T={\displaystyle \frac{1}{n}}{\displaystyle \sum _{n=1}^{n}f\left[V_T\left(x_i\right)\right]}$$

(9)

where Y_T denotes the mean of all decision trees, n is the number of decision trees, V_T(x_i) denotes the sub-datasets generated in Step 2; f denotes the bootstrap-based function.

The SPSSAU software https://spssau.net/?100000001 was employed to construct the random forest model, with specific parameter settings detailed in

Table 3. Random forest parameter settings.

Model	Parameter Name	Parameter Value
Random forest	Sample(breed) number	390
	Proportion of training set	0.8
	Number of decision trees	100
	Node split standard	Gini coefficient
	Minimum sample size for node Splitting	2
	Minimum number of samples for leaf node splitting	1
	Maximum depth of a tree	No limitation
	Maximum feature count limit	Automatically determined by the system
	If the replacement sampling be permitted	Yes
	If the out-of-bag data be tested	Yes

. Given the limited sample size, replacement sampling was used, permitting repeated selection of identical samples while allowing testing on unselected samples (out-of-bag data).

4. Results and Analysis

4.1. Results of the Indicator Evaluation System

Based on the results of the indicator evaluation system, box plots illustrating the development levels of China’s new energy system and its three targets are presented in Figure 5 and Figure 6. In the plots, the length of the box is determined by the 75th and 25th percentiles, reflecting data dispersion. A longer box indicates greater dispersion. The white point within the box denotes the median. The upper whisker indicates the maximum value, while the lower whisker indicates the minimum value.

As shown in Figure 5, during the study period, the overall development level of China’s new energy system displayed a steady upward trajectory. However, the median growth rate remained relatively low, below 0.5, indicating a comparatively slow pace of progress. Meanwhile, the increasing box length reveals intensifying spatial imbalances across regions. Despite this, the minimum value also rose, suggesting that underdeveloped regions were not neglected in energy system development. The most significant median increase occurred from 2016 to 2019, likely due to the implementation of China’s supply-side structural reform policy at the end of 2015. This policy involved large-scale restructuring of production capacity, including closing high-pollution, high-emission, low-efficiency enterprises, as well as upgrading the layout of resource-dependent regions and energy-intensive industries, all of which contributed to achieving the three objectives of the new energy system.

Based on Figure 6, the median development levels of the three objectives exhibit a gradient pattern: security > economy > green, with all three showing upward trends. The green objective consistently ranked lowest, reflecting the phased development characteristics typical of developing nations. After 2016, owing to supply-side reforms, the green objective improved significantly but plateaued after 2019, stabilizing at around 0.26–0.28. The box length for economic efficiency is substantially longer than that of the other two objectives, with the median located near the lower 25th percentile. This indicates pronounced spatial imbalance in energy utilization efficiency, with many regions showing low values and evidence of decoupling. However, some improvement has been observed since 2016. The security objective showed the most stable growth, with its upward trend accelerating in later years. Although dispersion increased, it remained the least dispersed among the three.

In summary, during the study period, China’s new energy system and its three objectives demonstrated a positive trajectory. Nevertheless, progress was relatively slow, and the overall absolute level remained low. Following the implementation of supply-side reforms, significant improvements were observed. A gradient pattern of security > economy > green was confirmed, with security showing the most stable development, the green objective stagnating after 2019, and the economic objective exhibiting the greatest imbalances.

4.2. Results and Analysis of the Random Forest Model

4.2.1. Model Robustness Testing

According to standard validation methods for random forest models, robustness was tested using MAE, MSE, and RMSE, while explanatory power was tested using R² and EVS [46]. The descriptions and test results for each indicator are presented in

Table 4. Model test results.

Indicators	Descriptions	Training Set	Testing Set
R2	Fit level, ranging from [0, 1], should be as close to 1 as possible.	0.885	0.834
MAE	Difference between the mean of actual values and the mean of fitted values, should be as close to 0 as possible.	0.007	0.017
MSE	Mean of the squared errors, should be as close to 0 as possible.	0.000	0.001
RMSE	MSE square root, average gap value	0.010	0.023
EVS	Measures the model’s explanatory power for data fluctuations, ranging from [0, 1], should be as close to 1 as possible.	0.885	0.836

.

As shown in Table 4, both the training and testing datasets exhibit R² and EVS values in the range of 0.83–0.89, indicating strong explanatory power. The minimal difference between training and testing R² values suggests that no overfitting occurred. With MAE, MSE, and RMSE all below 0.05, the model satisfies robustness requirements.

4.2.2. Assessment and Analysis of Driving Factors in All Regions

Based on the results of the random forest model, a feature weight diagram of the driving factors for the development of China’s new energy system is presented in Figure 7. The vertical axis represents the five driving factors, while the horizontal axis shows their feature weight values. Higher values indicate stronger driving effects, reflecting greater dependence of system development on that factor.

As shown in Figure 7, the feature weights of innovation fund input (FUN) and high-level labor input (HUM) are 0.42 and 0.27, respectively, both substantially higher than the other three factors. This demonstrates that China’s new energy system relies heavily on physical and human capital. In contrast, the feature weights of technology market size (MAR), share of tertiary industry (IND), and environmental regulation investment (ENV) are each close to 0.1, suggesting weak but synergistic effects.

Two main reasons explain this outcome. First, in the wake of supply-side reforms after 2015, China’s energy structure—especially the coal sector—experienced significant shocks. Policies such as capacity reduction, closure of outdated enterprises, stricter environmental standards, and relocation of high-energy-consuming clusters to less-polluted areas were implemented. While these measures optimized industrial structure and spatial distribution, they also temporarily slowed regional energy economic growth, deepening reliance on external capital. Second, China is still in the early stages of energy system transition, with insufficient infrastructure and technological tools. As a result, marginal returns on capital input are relatively high, strengthening the driving role of FUN. However, limitations in infrastructure have constrained the effectiveness of HUM, MAR, IND, and ENV. Consequently, the order of driving effects is FUN > HUM > (MAR ≈ IND ≈ ENV).

4.2.3. Accessing and Analysis of Driving Factors in Different Regions

To further explore regional disparities, random forest assessments were conducted separately for China’s four major economic regions (see Figure 1). Because the Northeast region comprises only three provincial-level units, its sample size was too small for independent modeling. Therefore, Northeast and Central China were combined due to similarities in economic foundations and industrial structures. Feature weight diagrams for different regions are presented in Figure 8, Figure 9 and Figure 10.

As shown in Figure 8, the results for Northeast and Central China display a polarized pattern compared with the national average. FUN and HUM exhibit significantly stronger driving roles, with a combined feature weight of 0.78, notably higher than in other regions. Conversely, MAR and IND show combined feature weights below 0.1, indicating minimal influence.

This is attributable to the heavy industrial base and resource concentration of Northeast and Central China. Supply-side reforms triggered more pronounced fluctuations in capacity and stagnation in local energy systems, further increasing reliance on external capital investment. Although these regions possess abundant reserves and strong industrial foundations, the “semi-planned market regulation” of China’s energy resource management system, combined with past irrational development models, has left enterprises lacking effective market-oriented tools. The energy market remains fragmented, limiting cross-regional coordination and technical exchange. Additionally, the dominance of traditional energy industries has created a crowding-out effect on the tertiary sector, weakening IND.

ENV shows a feature weight about two percentage points higher than the national average. This is largely due to the severe environmental degradation in these regions, where regulation can yield higher marginal benefits. However, constrained by weaker local economies, the intensity of environmental regulation remains insufficient, preventing ENV from exerting a stronger driving effect.

As shown in Figure 9, compared with Northeast and Central China, the feature weight values for FUN and HUM in East China decreased by 7 and 11 percentage points, respectively. This indicates that the development of the new energy system in the eastern region is advancing more rapidly, is less constrained by supply-side reforms, and shows a reduced dependence on capital. Nevertheless, FUN remains the dominant driving factor. Meanwhile, the feature weights for MAR and IND increased substantially, reaching 0.20 and 0.14, respectively, approaching the level of HUM. The reasons are twofold. On the one hand, as China’s economic hub, the eastern region benefits from a highly developed tertiary sector, advanced marketization, and strong capabilities in technological innovation and application. Building on phased achievements in new energy technologies such as photovoltaics [57], nuclear power [58], and automobile manufacturing [59], both MAR and IND exert stronger driving effects. On the other hand, under China’s regional policies, externalities from energy development and supply in eastern areas have been partially absorbed by other regions. Consequently, eastern energy systems face fewer conflicts between development and environmental governance, which decreases the phased role of ENV in this region.

As shown in Figure 10, the feature weights for FUN and HUM in West China are 0.31 and 0.29, respectively, indicating a nearly equal relationship. This demonstrates that the region’s new energy system is highly dependent not only on capital input but also on human capital, with both factors exerting synchronized driving effects. At the same time, MAR and ENV also show relatively notable driving effects, with their feature weight values being the closest to FUN and HUM among all regions. This suggests that, except for the limited contribution of IND due to the underdeveloped tertiary sector, the western region exhibits a relatively balanced distribution of driving factors. The primary reason is that the western region began its economic and social development later and relies heavily on the diffusion of fundamental elements from the eastern and central regions [60]. Its energy system has insufficient self-development capacity, and the tertiary sector remains underdeveloped, thereby producing a balanced but constrained driving structure.

5. Suggestions

Based on the literature and empirical findings, the main developmental issues of China’s new energy system can be summarized as follows:

Issue 1: Infrastructure development lags behind expectations, resulting in heavy reliance on capital investment and weakening the driving effects of technology markets and the tertiary sector.

Issue 2: Following the supply-side reform policies introduced at the end of 2015, temporary fluctuations emerged in production capacity, labor utilization, and economic contributions. The transition process became path-dependent on policy, which had limited effect on fostering technological innovation. Many enterprises reduced capacity through administrative measures rather than improving competitiveness via innovation, failing to generate endogenous momentum for technological advancement. This issue was particularly evident in Northeast and Central China.

Issue 3: The “island” phenomenon within regional energy systems remains severe. Development levels are unbalanced, standards are inconsistent, and cross-regional and cross-departmental coordination is weak. Technological effects are locked into specific locations, and fragmented markets hinder the rational allocation of fundamental resources.

Issue 4: Many regions in West, Northeast, and Central China continue to experience conflicts between environmental governance and economic development. Under the pressure of green transition policies, the shift from traditional to new energy systems has been marked by transitional difficulties, trade-offs, and mismatches between supply and demand.

Policy recommendations:

Accelerate integration of modern technologies. Promote the incorporation of 5G, artificial intelligence, and other modern technologies into the energy system. Leveraging their efficiency, intelligence, and virtualization can enable scientific layout and flexible scheduling of energy networks and control systems. This will shorten planning cycles, reduce trial-and-error costs, and expedite infrastructure construction for the new energy system.

Optimize market mechanisms to achieve the three development targets.

(1) Improve the carbon emissions trading market, clarify functional roles of different regions, and prevent homogeneous development that wastes capital and neglects regional comparative advantages. This will also help mitigate conflicts between environmental governance and economic development.

(2) Fully utilize market mechanisms to regulate energy prices. Moderately widen time-of-use electricity price differentials and rely on price signals to guide society in adjusting electricity demand across different periods, thereby alleviating supply-demand mismatches and security risks in the power system.

(3) Utilize government market intervention tools to establish a nationwide energy system interaction platform, promoting integrated development of energy markets. This includes creating trading systems for coal, oil and gas futures, and new energy, gradually dismantling cross-regional, cross-sectoral, and cross-level barriers within the energy system.

Finally, efforts should be intensified to promote the coordinated development of the energy industry chain, energy markets, and related service sectors. The key lies in restoring the commodity attributes of energy, leveraging energy markets and financial services to enhance the capital accumulation capacity of the energy system. Once sufficient capital accumulation is achieved, service industries such as electronic information and science and technology can help lower barriers to innovation. This process will gradually establish a virtuous cycle characterized by reduced reliance on external capital, shorter technological innovation cycles, expansion of technology market size, and strengthened developmental momentum in the tertiary sector.

6. Discussion

This article evaluates the development level of the new energy system in 30 provincial-level administrative regions of China from 2011 to 2023, analyzes its driving factors, and, based on empirical results, discusses the development status, driving mechanisms, and key issues while providing policy recommendations. Compared with previous research models that primarily relied on qualitative approaches, this study establishes a quantitative framework for analyzing new energy systems through think tank research techniques, indicator system evaluation, and machine learning.

On the one hand, by summarizing perspectives from existing literature and integrating the “impossible triangle” theory, this study proposes a conceptual framework for constructing an indicator evaluation system for China’s new energy system centered on three core targets: green, economy, and security. On the other hand, based on the resource, economic, commodity, and industrial attributes of energy—along with an emphasis on environmental considerations—this study identifies driving factors that capture the distinctiveness of the “new” energy system. To overcome the limitations of traditional linear regression models in analyzing composite dependent variables, a random forest model was employed to evaluate the contribution of each driving factor. In this way, this article not only expands the research perspective but also enriches methodological approaches in the field.

Nonetheless, this study has three limitations. First, the scope of the new energy system is broad, and due to data constraints, only the most representative indicators were selected. The indicator system therefore still has room for optimization; for example, by expanding the criterion layer or applying more granular classification across layers, efficiency could be enhanced. Second, the selected driving factors are rooted in the domestic context of China, without incorporating systematic analysis of external drivers such as foreign direct investment [61], international energy cooperation [62], or the energy policies of neighboring countries [63]. Future research should incorporate these external factors. Third, given the temporal synchrony between the empirical data and China’s innovation incentive policies, the driving factors selected may exhibit interactions with policy effects. However, this study cannot rigorously distinguish between these factors. Further research could build upon this work by employing policy evaluation models such as the difference-in-differences (DID) model to explore causal mechanisms more accurately.

Notably, both the data and perspective of this study focus on the macro level. The development of new energy systems undoubtedly requires more microlevel and technical research to support this study. For example, to increase the stability of the supply and demand for integrated energy systems (IES) in the face of cyberattack threats, Li et al. proposed an innovative model-free resilience scheduling method based on state-adversarial deep reinforcement learning for integrated demand response-enabled IES [64]. Huang et al. worked to increase the operational safety of an integrated electricity-gas system (IEGS) and proposed a reduced-order transfer function model that considers an electric-driven compressor for accurate and efficient dynamic analysis of the IEGS [65]. These studies provide valuable counterpoints to the main arguments of this study and contribute to broadening the research perspectives on new energy systems.

Future research directions should focus primarily on two aspects. On the one hand, the indicator evaluation system constructed in this study is based on the impossible triangle theory. While it partially overlaps with common concepts reflecting energy system characteristics—such as sustainability, affordability, and reliability—it does not fully encompass them. Future research can focus on improving the evaluation framework for China’s new energy system in three ways. On the other hand, the empirical findings of this article highlight significant spatial imbalances in the development of China’s new energy system. This implies that spatial-geographical factors inevitably exert an additional influence. Future studies could adopt GIS technology or spatial econometric models to further explore the role of spatial effects. For example, Zhong et al. employed spatial feature analysis and a random forest model to conduct a comprehensive analysis of the spatiotemporal patterns and influencing factors of the coordinated development of China’s population-ecology-energy-digital economy (PEED) system [66], providing a reference for this approach. Moreover, during spatial effect analysis, the research scale can be further refined to explore potential heterogeneity within provinces.

7. Conclusions

The contributions of this article can be summarized as follows: First, based on the framework of the Analytic Hierarchy Process, this study constructed an indicator evaluation system by integrating literature perspectives through text-mining of think tank research. Using the “impossible triangle” theory, three core development targets—green, economy, and security—were identified. Second, the entropy weight method was employed to evaluate the development level and characteristics of China’s new energy system across 30 provincial-level regions from 2011 to 2023.

Drawing on the intrinsic attributes of energy, the study identified key driving factors and applied a random forest model to systematically analyze their effects, providing policy-oriented recommendations.

The main innovations of this article are as follows:

A quantitative analytical model for new energy systems was proposed, addressing the paucity of quantitative research in this field.

A framework for constructing an indicator evaluation system that integrates think tank technology, the AHP, and the Entropy Weight Method was proposed. By utilizing the noun recognition capabilities of text mining technology, it maximizes the extraction of implicit information from the academic literature, thereby further enhancing the objective goals of the AHP and the Entropy Weight Method. This approach offers new insights into multidimensional energy system evaluation methods for other emerging economies.

A random forest model reflecting “new” driving factors in the innovation process of energy systems was developed, enhancing scientific rigor while circumventing the shortcomings of traditional regression approaches.

The main conclusions are as follows:

During the study period, China’s new energy system and its three core targets exhibited a consistent upward trajectory. However, the overall pace was relatively slow, starting from a low baseline, and accompanied by significant spatial imbalances. Following the implementation of supply-side structural reform policies, development improved noticeably. Yet, in some regions, inherent structural issues were exacerbated. The three targets followed a gradient pattern: security > economy > green. Among these, security showed the most stable progress, the green target stagnated after 2019, and the economic target was most affected by uneven regional development.

Innovation funding and high-level labor input were the dominant driving factors, while technology market size, tertiary sector development, and environmental regulation investment acted as supplementary drivers, with clear regional variations. Northeast and Central China exhibited pronounced dependence on capital inputs, forming a polarized driver pattern. East China demonstrated strong driving roles for technology market size and tertiary sector development, while environmental regulation investment played the smallest role due to favorable baseline conditions and external compensation effects. West China, with its later start and dependence on factor diffusion from central and eastern regions, showed a relatively balanced driving pattern, though constrained by an underdeveloped tertiary sector.

These findings reveal the phased challenges of China’s new energy system, including inadequate infrastructure, weak system coordination, underdeveloped market mechanisms, and persistent difficulties in balancing regional development with environmental governance. To address these issues, targeted corrective measures must be urgently formulated to ensure coordinated, sustainable, and innovation-driven energy system transformation.