Impact of Zero-Waste City Pilot Policies on Urban Energy Consumption Intensity: Causal Inference Based on Double Machine Learning

Abstract

To scientifically assess the energy-saving effects of China’s zero-waste city pilot (ZWCP) policies and provide empirical evidence and policy insights for advancing pilot policies and accelerating energy conservation and emission reduction goals, this study selected 274 cities in China from 2010 to 2022 as the research sample, employing a double machine learning model to empirically analyze the impact of pilot policies on urban energy consumption intensity. The research results demonstrate that the ZWCP policies significantly reduced energy consumption intensity in pilot areas. Channel analysis reveals that this policy exerted a restraining effect on energy consumption intensity through industrial structure upgrading, green technology innovation, and enhanced environmental awareness. Heterogeneity analysis shows that policy effects were more pronounced in non-urban agglomeration regions, inland areas, and small-to-medium-sized cities. This study provides crucial decision-making references for the promotion and implementation of ZWCP policies during the “14th Five-Year Plan” period.

1. Introduction

Urbanization is progressing at an unprecedented pace. The United Nations predicts that by 2050, the global urban population will increase by 2.2 billion compared to 2021, with the urbanization level rising from 56 to 68%. Urbanization, as a crucial driver of economic growth, simultaneously presents severe environmental challenges [1]. Against this backdrop, the zero-waste city concept has emerged as an innovative governance model addressing ecological dilemmas in urban development. Urban solid waste management not only concerns environmental quality but also serves as a critical lever for achieving carbon peak and carbon neutrality goals [2]. With continuous urban population expansion, fossil fuel consumption surges, leading to a dramatic increase in greenhouse gas emissions and exacerbating urban climate change. Synchronous industrialization during the urbanization process further intensifies the generation of pollutants such as exhaust gases and industrial waste [3].

Within the complex ecological challenges of urban development, energy consumption intensity emerges as a key measurement indicator and improvement focus. The 2024 Chinese Government Work Report clearly stated the importance of bolstering the construction of ecological civilization, encouraging the development of green and low-carbon initiatives, and consistently moving toward carbon peaking and carbon neutrality, aiming for a roughly 2.5% decrease in energy consumption per unit of GDP [4]. This signifies that reducing energy consumption intensity and achieving economically efficient energy conservation have become hard constraints in government economic work. Statistics reveal that in 2020, urban energy consumption accounted for 86.9% of China’s total national energy consumption, approximately 18% higher than the international average [5]. Consequently, energy consumption has become primarily concentrated in cities, and effectively reducing urban energy consumption intensity is of paramount significance for comprehensive energy conservation in economic and social domains.

To achieve economic energy-saving objectives, the State Council’s General Office initiated the zero-waste city pilot (ZWCP) program in 2019. According to data from the Ministry of Ecology and Environment, the first batch of “11 + 5” ZWCP projects has been officially launched, aiming to explore new pathways for urban sustainable development through source reduction, resource utilization, and the safe disposal of solid waste. Although urban solid waste management accounts for merely 3–5% of total societal carbon emissions, source reduction, resource recycling, and high-standard harmless disposal can significantly decrease greenhouse gas emissions. Zero-waste city construction is not only an innovative concept in solid waste management but also a crucial mechanism for achieving carbon peak and carbon neutrality goals. Pilot cities, based on regional industrial structures and development stages, precisely identify critical points in the solid waste lifecycle, and through systematic integration and collaborative linkage, continuously enhance urban solid waste reduction, resource utilization, and harmless treatment levels [6].

The existing literature on energy consumption intensity primarily focuses on the following aspects—First, research on factors influencing energy consumption intensity: Scholars have explored key elements affecting energy consumption intensity based on multiple dimensions, including industrial structure, technological innovation, and economic development. Li et al. [7] found that low-carbon city pilot policies significantly reduce urban energy intensity through green technological innovation and environmental governance, with the mediation effect of green technological innovation reaching 40.80%. Khosravi et al. [8] demonstrated that the environmental “fee-to-tax” policy can reduce energy consumption by optimizing industrial structure and increasing technological innovation investment. Second, research on the impact mechanism of environmental policies on energy consumption intensity: Current studies predominantly examine the impact of various types of environmental policies on energy savings and emission reductions. For instance, Liao et al. [9] studied the spatial energy-saving effects of energy use rights trading systems, while Wang et al. [10] analyzed energy-saving and emission reduction mechanisms from a digital economy perspective. These studies indicate that environmental policies are crucial means of regulating energy consumption intensity. Third, regional heterogeneity research: Scholars have discovered significant regional differences in the energy-saving effects of environmental policies. Liu et al. [11] noted that low-carbon city pilot policies demonstrate more pronounced energy-saving effects in eastern and western cities, southern cities, and top 100 economic cities. Shu et al. [12] similarly showed that environmental taxes have more significant energy consumption reduction effects in eastern regions and old industrial areas. However, compared to other environmental policy research, academic studies on ZWCP policies remain relatively weak: Bi et al. [13] used DID and machine learning methods to find that ZWCP policies significantly advance green technological innovation; Ray et al. [14] employed quasi-experimental methods to confirm the policy’s effectiveness in promoting urban green and low-carbon transformation; and Liu et al. [15], based on microenterprise-level data, further revealed the significant positive impact of ZWCP policies on enterprise green innovation.

Existing research still has the following limitations: firstly, research on the impact of ZWCP policies on energy consumption intensity is relatively scarce; secondly, existing studies primarily focus on single policies such as low-carbon cities and environmental taxes, lacking systematic analysis of ZWCP policies; thirdly, there is a lack of comprehensive research on the action mechanisms of and regional variations in ZWCP policies; and fourthly, traditional econometric methods may face endogeneity and omitted-variable issues, impacting the reliability of causal inference.

Based on this, this study selects 274 cities in China from 2010 to 2022 as the research sample and employs a double machine learning model to systematically examine the impact of pilot policies on urban energy consumption intensity. The study focuses on the following key questions: Do ZWCP policies reduce energy intensity in pilot cities? What are the impact mechanisms of pilot policies on urban energy intensity? What heterogeneous characteristics exist in pilot cities’ effects on energy intensity? By deeply investigating these issues, this paper aims to clarify the implementation effects of China’s ZWCP policies, scientifically assess the energy-saving effects of pilot policies, and provide a decision-making basis for the promotion of ZWCP policies during the “14th Five-Year Plan” period and the achievement of the “30·60” carbon emission target.

The marginal contributions of this paper are primarily reflected in three aspects: First, this study represents an initial comprehensive examination of how ZWCP policies affect urban energy consumption intensity, including their impact effects, mechanisms of action, and regional differences. This research contributes to the theoretical framework surrounding environmental policies and urban sustainable development. By constructing multi-dimensional theoretical analysis paths, this paper reveals the complex action mechanisms of urban energy consumption intensity, providing a new academic perspective for understanding the micro-level impacts of environmental regulation. Second, by introducing a double machine learning model for empirical analysis, this paper effectively controls covariate influences and overcomes the “curse of dimensionality” in traditional econometric methods. Compared to traditional econometric methods, this model significantly improves the accuracy and reliability of causal inference, providing a more precise technical approach for quantitative research on environmental policy impacts. Finally, based on machine learning technology, this paper uses multiple programming languages to construct robust tests for causal forest models and visualizes the importance of control variables, effectively avoiding the inherent defects of machine learning’s “black box” operations. This multi-dimensional, multi-tool research paradigm not only enhances the reliability of empirical results but also provides a reference for subsequent environmental policy research.

2. Policy Background and Research Hypotheses

2.1. Policy Background

With the in-depth advancement of sustainable development concepts, the social development vision of “zero waste” gradually became a consensus in the international community during the early 21st century. Countries have successively promoted this concept through legislation and policies: Japan pioneered the “Basic Law for Promoting a Recycling-Oriented Society” in 2000 and passed the “Fourth Basic Plan for Promoting a Recycling-Oriented Society” in 2019, setting seven national initiatives and specific targets for 2025; the European Union launched the “Towards a Circular Economy: European zero waste Program” and “Circular Economy Package” in 2014 [16]; Singapore clearly proposed the development vision of becoming a “zero-waste” nation in the “Singapore Sustainable Blueprint 2015”. This concept has also received an active response at the urban level, with cities like San Francisco, Vancouver, and Stockholm successively developing “zero-waste city” blueprints. More notably, 23 member cities of the C40 Cities Group jointly signed the “Towards Zero Waste Declaration” [17], emphasizing that zero-waste cities are an inevitable choice for sustainable, prosperous, and livable cities in the future and setting a quantitative goal of reducing 87 million tons of waste by 2030. These practices fully demonstrate that promoting the construction of “zero-waste cities” has become an essential path to driving sustainable urban economic and social development and solving solid waste pollution problems [18].

ZWCP construction in China began in early 2018, with a strategy aimed at minimizing urban solid waste generation, maximizing resource utilization, and ensuring safe disposal. In June 2018, the Central Committee of the Communist Party of China and the State Council included the ZWCP in the “Opinion on Comprehensively Strengthening Environmental Protection and Resolutely Winning the Battle of Pollution Prevention”, as an important component of the soil protection campaign. In December 2018, the State Council Office issued the “Work Plan for Zero-waste City Pilot Construction”, planning to select approximately 10 pilot cities. In April 2019, the “11 + 5” pilot cities and regions, represented by Shenzhen and Baotou, officially launched ZWCP reform, including 11 pilot cities and 5 pilot special zones (including industrial parks and counties). In November 2021, to further promote the efficient implementation of garbage classification, the Ministry of Ecology and Environment, together with 17 other departments, issued the “Several Opinions on Further Promoting Domestic Waste Classification Work”. In December of the same year, they also released the “Work Plan for Promoting Zero-waste City Construction during the 14th Five-Year Plan Period”, positioning ZWCP construction as a key measure to achieve carbon peaking and carbon neutrality, developing a specific indicator system across urban spatial and temporal scales, and providing crucial support for achieving the modernization goal of harmonious coexistence between humans and nature. The names of ZWCP areas are shown in

Table 1. ZWCP areas.

Categories	Name
Pilot cities	Baotou, Chongqing, Panjin, Ruijin, Sanya, Shaoxing, Shenzhen, Tongling, Weihai, Xuchang, Xining, Xuzhou
Pilot special zones	Beijing Economic and Technological Development Area, China–Singapore Tianjin Eco-City, Guangze, Xiong’an New Area

, and Figure 1 displays the distribution of ZWCP experimental zones.

2.2. Research Hypotheses

This study aims to investigate the impact of ZWCP policies on urban energy consumption intensity, analyzing the mediating roles of industrial structure upgrading, green technological innovation, and environmental awareness enhancement among the government and investors. As an innovative environmental governance model, ZWCP policies primarily target effective reduction in energy consumption intensity by promoting resource-efficient utilization and waste minimization. According to Li et al. [19], this policy guides enterprises to adopt advanced green technologies and construct circular, low-carbon production and living modes by promoting solid waste reduction, resource recycling, and harmlessness, which not only effectively improves total-factor carbon emission efficiency but also facilitates urban green transformation. Existing research indicates that low-carbon city pilot policies can significantly reduce regional energy intensity by promoting green technological innovation and optimizing industrial structure [20], providing theoretical foundations for the potential benefits of ZWCP policies in resource management and environmental protection. Based on this, we propose Hypothesis 1.

Hypothesis 1 (H1). 

ZWCP policies can significantly reduce urban energy consumption intensity.

Moreover, ZWCP policies can indirectly reduce urban energy consumption intensity by promoting industrial structure optimization. According to Wang et al. [21], industrial structure upgrading can guide production factors toward more efficient green industries, restrict and eliminate backward high-energy-consumption and high-pollution capacities, and promote industrial structure transformation in a green and low-carbon direction. Research demonstrates a significant negative correlation between industrial structure adjustment and energy consumption intensity. Through the optimization of secondary industry structure, energy efficiency can be effectively improved and energy consumption reduced [22]. Therefore, we propose Hypothesis 2.

Hypothesis 2 (H2). 

ZWCP policies reduce urban energy consumption intensity through industrial structure upgrading.

ZWCP policies can also significantly reduce urban energy consumption intensity by promoting green technological innovation. Based on Zhang et al. [23], the policy’s role in green technological innovation manifests through the establishment of specialized technical support systems, focusing on zero growth of industrial solid waste and providing data and technical support for green technological innovation through informatized management and technological upgrades, facilitating information sharing and collaboration among the government, enterprises, and research institutions. Existing research points out that technological innovation can not only improve production efficiency but also effectively reduce energy consumption per unit of product [24]. Accordingly, we propose Hypothesis 3.

Hypothesis 3 (H3). 

ZWCP policies reduce urban energy consumption intensity through green technology innovation.

Finally, ZWCP policies can indirectly reduce urban energy consumption intensity by enhancing environmental awareness among the government and investors. According to Qin et al. [25], the mechanisms of environmental awareness enhancement include the government strengthening environmental governance, incorporating ZWCP construction achievements into performance assessments, providing financial support and green financial innovations to enterprises, enhancing public supervision of corporate environmental behavior, and motivating enterprises to proactively fulfill environmental responsibilities. Research indicates that improved environmental awareness can significantly influence corporate decision-making and drive more environmentally friendly production methods [26]. Therefore, we propose Hypothesis 4.

Hypothesis 4 (H4). 

ZWCP policies reduce urban energy consumption intensity through fostering enhanced environmental awareness.

Figure 2 illustrates the research hypothesis framework.

3. Research Design

3.1. Model Specification

To study the impact of the ZWCP policies on energy consumption intensity and addresses the “dimensionality curse” that arises under high-dimensional control variables, this paper employs a double machine learning (DML) framework to identify causal effects. The DML method is used to estimate the partially linear model, and through “Neyman orthogonalization” and “cross-fitting”, it ensures that the estimated parameters are robust.

3.1.1. Basic Model Specification

Incorporating the discussion by Kleinberg et al. [27] on the application of machine learning in econometrics, we set the control variable $X_i$ as potentially observable and unobservable factors, defined as follows:

$$EC{I}_i={\theta }_{0}Even{t}_i+g\left(X_i\right)+U_i$$

(1)

$$Even{t}_i=m\left(X_i\right)+V_i$$

(2)

where $EC{I}_i$ represents the energy consumption intensity, $Even{t}_i$ indicates the ZWCP policy’s implementation changes (with 1 for implementation and 0 for non-implementation), $X_i$ is the vector of control variables, $g(\cdot )$ is an unknown function, $U_i$ is the error term, and ${\theta }_0$ is the parameter of interest.

3.1.2. Neyman Orthogonalization Principle

With reference to Chernozhukov et al. [28], when utilizing machine learning for causal inference, it is necessary to ensure the identification of the parameter ${\theta }_0$. To this end, we introduce Neyman orthogonalization, constructing the following “orthogonal number”:

$${\psi }_i(\theta ,g,m)=\left(Even{t}_i-m\left(X_i\right)\right)\left[\left(EC{I}_i-g\left(X_i\right)\right)-\theta \left(Even{t}_i-m\left(X_i\right)\right)\right]$$

(3)

If $\theta ={\theta }_0$ and the machine learning model is approximately equal to $g\left(X_i\right)$ and $m\left(X_i\right)$,

$$E\left[\psi \left(\theta ,g\left(X_i\right)\right)\right]=0$$

(4)

Then, we have

$$E\left[Even{t}_i-m\left(X_i\right)\right]\left(EC{I}_i-g\left(X_i\right)\right)-{\theta }_{0}E\left[Even{t}_i-m\left(X_i\right)\right]=0$$

(5)

By solving this, we obtain

$${\theta }_0={\displaystyle \frac{E\left[Even{t}_i-m\left(X_i\right)\right][EC{I}_i-g\left(X_i\right)]}{E\left[Even{t}_i-m\left(X_i\right)\right]}}$$

(6)

To calculate the expected value, we use the sample average as a substitute and estimate the function values using machine learning:

$$\widehat{g}\left(X_i\right)\approx g\left(X_i\right),\widehat{m}\left(X_i\right)\approx m\left(X_i\right)$$

(7)

Thus, the estimated regression form is

$${\widehat{\theta }}_0={\left[{\displaystyle \frac{1}{N}}\sum _{i=1}^N{\left(Even{t}_i-\widehat{m}\left(X_i\right)\right)}^2\right]}^{-1}\cdot \left[{\displaystyle \frac{1}{N}}\sum _{i=1}^N\left(Even{t}_i-\widehat{m}\left(X_i\right)\right)\left(EC{I}_i-\widehat{g}\left(X_i\right)\right)\right]$$

(8)

Through this “normalization” design, it is possible to reduce the bias caused by small factor estimates, allowing for high-dimensional control conditions to be satisfied. The estimation of the relevant parameter ${\theta }_0$ becomes more reliable (provided that certain conditions regarding convergence speed and stability are met), which allows us to achieve $\sqrt{N}$ convergence and improve consistency.

3.1.3. “Third Term” Bias and Cross-Fitting

Following Belloni et al. [29], if machine learning training (estimation) and residual regression of the target parameter are performed simultaneously on the same sample batch, a “Third Term” bias will emerge, causing endogenous interference in the estimation of the parameter. To address this, cross-fitting can be used to further eliminate this bias: randomly divide the sample into several folds (e.g., five folds), then iteratively perform machine learning fitting of the parameters in the training set, and calculate residuals in the corresponding test set.

$$\widehat{ECI}=ECI-\hat{g}\left(X_i\right),\widehat{Event}=Event-\widehat{m}\left(X_i\right)$$

(9)

By substituting this into Equation (8), we obtain ${\widehat{\theta }}_0$, which ultimately averages the results of various estimates to achieve the overall estimate ${\widehat{\theta }}_0$. Through this process, “functional estimation” and “core parameter estimation” are completed for different samples, ensuring independence and effectively avoiding high-dimensional estimates leading to external bias.

In conclusion, this research utilizes machine learning techniques to analyze the effect of ZWCP policies on energy consumption intensity. The model integrates the relationship between energy consumption and various factors, primarily reflecting the following aspects: The method allows for the representation of $g(\cdot )$ and $\mathrm{m}(\cdot )$ as smooth functions that can capture the complexity of the relationship between energy consumption and its determinants. By combining machine learning techniques (such as deep learning networks, Lasso, etc.), it can effectively capture the characteristics of the relationship. In estimating the smooth functions, the Neyman orthogonalization method is utilized to ensure the accuracy of the estimates and to maintain the independence of the errors in the machine learning framework, thereby ensuring the reliability of the parameter ${\theta }_0$. In this effective framework, this method can achieve convergence at a rate of $\sqrt{N}$ and can effectively maintain the corresponding bias.

3.2. Variable Definitions

3.2.1. Dependent Variable

The dependent variable in this study is energy consumption intensity (ECI). With the increasing maturity of remote sensing technology, the high-precision and broad coverage characteristics of satellite data provide reliable technical support for calculating energy consumption intensity at the county level. Following the research methodology of Li et al. [30], this paper constructs an energy consumption intensity indicator through the ratio of total energy consumption to regional GDP, which directly reflects the efficiency of energy utilization in regional economic activities.

Considering the precision issues of the model at different scales, this paper adopts a linear model:

$$E_{it}=k_{t}D{N}_{it}$$

(10)

where $E_{it}$ represents the total energy consumption of province $i$ in year $t$; $k_t$ is a coefficient for year $t$; and $D{N}_{it}$ is the number of categories of ashes in province $i$ in year $t$. Through the use of ArcGIS 10.0, the model is established based on the energy consumption data of Chinese cities, calculating the energy consumption scale for each city and ultimately constructing the total energy consumption for cities in China.

3.2.2. Core Independent Variable

The primary independent variable in this research is the ZWCP policy (Event). In January 2019, the General Office of the State Council released the “Work Plan for the Zero-Waste City Construction Pilot Program”, which was jointly initiated by 18 departments, including the Ministry of Ecology and Environment. This paper considers this policy as a quasi-natural experiment and applies double machine learning techniques to assess its effects. Specifically, a dummy variable is constructed based on the “zero-waste city construction pilot list” published by the Ministry of Ecology and Environment in 2019, where the value is set to 1 for cities that are part of the pilot program after its implementation and 0 for those that are not.

3.2.3. Control Variables

The double machine learning method can automatically screen preselected control variables to obtain control variable combinations with higher prediction accuracy, thereby avoiding problems such as control variable redundancy. Therefore, to ensure the scientific validity of the policy evaluation, with reference to the research of Chen et al. [31] and Wang et al. [32], this study selects as many control variables as possible that may affect energy consumption intensity.

① Economic Development (Gdp): The economic development level is one of the important factors affecting energy consumption intensity. As the economy develops, increased economic development levels lead to higher total energy consumption, but due to technological progress and industrial structure optimization, energy consumption per unit of GDP tends to decrease. This indicates a close relationship between regional GDP growth and improved energy utilization efficiency. This paper uses the natural logarithm of regional GDP to measure the economic development level.

② Population Density (Density): The previous literature has indicated that population and industrial agglomeration lead to technological progress and improve overall energy efficiency, thereby reducing energy consumption intensity. In this paper, we use the natural logarithm of the year-end total urban population to represent population density.

③ Internet Development (Internet): The Internet can influence energy consumption intensity by improving information transmission efficiency and optimizing resource allocation. On the one hand, the popularization of Internet technology has promoted new work and lifestyle patterns such as remote work and online education, reducing energy consumption in traditional scenarios. On the other hand, the Internet enables smart manufacturing and refined management, helping enterprises achieve the digital transformation of production processes, improving energy utilization efficiency and thereby reducing energy consumption intensity. This paper uses the natural logarithm of international Internet users per 10,000 people to represent the Internet development level.

④ Transportation Development (Transport): Transportation can affect regional industrial layout and energy consumption patterns, potentially impacting energy consumption intensity. This paper uses the natural logarithm of highway passenger volume to represent the transportation development level.

⑤ Foreign Trade Openness (Fdi): Differences in production factor endowments between countries lead to the global division of labor and trade. Economic globalization promotes the free flow of global goods, optimizes resource allocation, accelerates technology transfer, and promotes industrial structure adjustment. However, it also transfers energy consumption in goods production through international trade, changing the spatial–temporal distribution of global energy consumption intensity. This paper uses the ratio of regional total imports and exports to regional GDP to measure the degree of foreign trade openness.

⑥ Industrial Structure (Structure): Different industrial structures have different energy demands, and per capita energy consumption varies with industrial structure changes. Meanwhile, industrial structure also affects development strategy choices; generally, economies with lower-level industrial structures are more likely to choose catch-up strategies. This paper uses the ratio of secondary- to tertiary-industry added value to measure the industrial structure level.

⑦ Urbanization (Urban): On the one hand, urbanization can affect per capita energy consumption through the economies of scale produced by agglomeration effects. On the other hand, since development strategies that violate comparative advantages will suppress urbanization levels, conversely, higher urbanization levels indicate a lower likelihood of governments choosing development strategies that violate comparative advantages, i.e., a lower TCI index. This paper uses the proportion of the year-end urban population to the total population to represent the urbanization level.

⑧ Education Level (Edu): The education level reflects improvement in human capital and has important impacts on regional economic development and environmental governance capabilities. Higher education levels can enhance environmental awareness among the public and enterprises, thereby reducing energy consumption intensity. This paper uses the natural logarithm of college students per 10,000 people to measure the region’s education level.

Considering the potential complex nonlinear relationships among variables, following Jiang [33], we incorporate quadratic terms of continuous variables within the double machine learning framework. This approach not only effectively handles high-dimensional control variables but also automatically selects the most relevant predictors, thereby enhancing estimation accuracy. Additionally, to control for unobserved heterogeneity, we include both year and individual fixed effects in our model.

3.3. Sample Selection and Data Sources

This study systematically examines the impact of the ZWCP policy on energy consumption intensity in China using panel data from 274 prefecture-level cities spanning 2010–2022. Given that municipalities (Beijing, Shanghai, Tianjin, and Chongqing) possess distinctive characteristics in terms of economic development and innovation capacity that might affect the reliability of baseline regression results, these four cities are excluded from our sample. For data processing, we employ linear interpolation to fill in partial missing values and remove city samples with high proportions of missing data. Additionally, we apply 1% winsorization at both tails for continuous variables to mitigate the influence of outliers. The measurement of total energy consumption is based on night-time light data from the Defense Meteorological Satellite Program (DMSP). The identification of ZWCP cities follows the official list published by the Ministry of Ecology and Environment in 2019. Other city-level data are sourced from the China Research Data Services Platform (CNRDS), China Stock Market & Accounting Research Database (CSMAR), and China City Statistical Yearbook.

Table 2. Descriptive statistics.

Var	Obs	Mean	SD	Min	Max
ECI	3230	0.786	0.446	0.044	4.513
Event	3230	0.107	0.309	0.000	1.000
Gdp	3230	16.599	0.909	14.177	19.596
Density	3230	5.755	0.900	2.833	7.200
Internet	3230	2.923	0.746	0.011	5.246
Transport	3230	8.163	1.145	2.398	12.566
Fdi	3230	0.015	0.015	0.000	0.068
Structure	3230	1.215	0.614	0.187	9.196
Urban	3230	0.463	0.177	0.096	1.000
Edu	3230	4.704	1.138	2.197	7.216

presents the descriptive statistics of key variables.

4. Empirical Analysis

4.1. Baseline Estimation Results

This study utilizes a partially linear double machine learning model to assess the impact of the ZWCP policy on urban energy consumption intensity. In the experimental design, we implemented a 1:4 sample splitting ratio and applied random forest algorithms for cross-fitting estimation of both the main and auxiliary regressions, with detailed findings displayed in

Table 3. Baseline regression results.

	(1)	(2)	(3)
Var	ECI	ECI	ECI
Event	−0.044 **	−0.042 ***	−0.039 ***
	(0.021)	(0.015)	(0.014)
Control (single term)	N	Y	Y
Control (quadratic term)	N	N	Y
City FE	Y	Y	Y
Year FE	Y	Y	Y
N	3230	3230	3230

Notes: The dependent variable is energy consumption intensity. Column (1) includes city fixed effects and year fixed effects. Column (2) adds linear terms of control variables. Column (3) adds quadratic terms of control variables. Robust standard errors in parentheses. *** p < 0.01; ** p < 0.05.

. The empirical analysis was performed at three levels: First, in the baseline model, which controls solely for city and time fixed effects (Column 1), the estimated coefficient for the core explanatory variable is −0.044, significant at the 5% level, which tentatively validates that the ZWCP policy significantly inhibits urban energy consumption intensity. Second, to mitigate potential omitted-variable bias, we progressively added linear terms (Column 2) and quadratic terms (Column 3) of control variables to the baseline model. The results indicate that across all model specifications, the estimated coefficient of the core explanatory variable—the event—remains robustly negative and significant at the 1% level, providing strong evidence that the ZWCP policy can significantly decrease energy consumption intensity in the pilot areas, thus confirming Hypothesis 1 of this study.

Meanwhile, to better understand the importance of control variables in the double machine learning process, we generated a feature importance plot using random forests (see Figure 3). We primarily relied on Python’s (v.3.12) scikit-learn library for random forest training and cross-validation, while visualization was accomplished through matplotlib. Specifically, we employed RandomForestRegressor in conjunction with KFold to execute the cross-validation process in double machine learning and utilized matplotlib.pyplot to generate convergence plots of fold-specific and aggregate coefficient estimates. All parameters were maintained at their default settings from Stata’s DDML package (v.18) to ensure the reproducibility and comparability of results. The figure clearly demonstrates that among all first-order terms, GDP exhibits the strongest explanatory power for urban energy consumption intensity, followed by Internet penetration rate (Internet), industrial structure (Structure), and population density (Density). This suggests that the level of economic development, degree of informatization, and industrial structure adjustment influence urban energy consumption intensity to varying degrees.

4.2. Robustness Checks

4.2.1. Adding New Fixed Effects

To comprehensively examine the dynamic characteristics of cities over time and mitigate potential omitted-variable bias, we incorporated city–year fixed effects into our baseline model. Given that cities within the same province in China’s administrative system often exhibit similarities in policy implementation and geographical features, we additionally included province–year fixed effects as supplementary controls [12]. Columns (1) and (2) in

Table 4. Robustness check I.

	(1)	(2)	(3)	(4)	(5)
Var	City–Time Trends	Province–Time Trends	LCC	CT	SC
Event	−0.041 ***	−0.039 ***	−0.034 **	−0.036 **	−0.034 **
	−0.014	−0.013	−0.015	−0.015	−0.015
Control (single term)	Y	Y	Y	Y	Y
Control (quadratic term)	Y	Y	Y	Y	Y
City FE	Y	Y	Y	Y	Y
Year FE	Y	Y	Y	Y	Y
N	3230	3230	3230	3230	3230

Notes: The dependent variable is energy consumption intensity. Columns (1) and (2) employ new fixed effects. Columns (3) to (5) include policy dummy variables for Low-Carbon City (LCC), Emissions Trading (ET), and Smart City (SC) pilots. Robust standard errors are reported in parentheses. *** p < 0.01; ** p < 0.05.

present the regression results after incorporating these two types of fixed effects. The findings demonstrate that even after controlling for these fixed effects, the negative impact of the ZWCP policy on energy consumption intensity remains significant, further validating the robustness of our research conclusions.

4.2.2. Excluding the Impact of Other Policies

During the same period, other relevant policies may have affected energy consumption intensity in pilot areas, potentially interfering with the identification of ZWCP effects. To address this concern, following Zeng et al. [34], this study controls for three concurrent policies: the Low-Carbon City pilot policy, the Carbon Trading pilot policy, and the Smart City pilot policy. The selection of these policies is based on two considerations: First, the zero-waste city initiative, as an environmental policy, is inherently connected to policies related to carbon emission reduction and trading. Second, improving solid waste treatment efficiency relies on the Smart City policy’s role in enhancing urban management efficiency and promoting industrial upgrading. Accordingly, this study incorporates these three policies as dummy variables in the double machine learning model as covariates. Columns (3) to (5) in Table 4 present the corresponding results. The findings reveal that although the regression coefficients of the ZWCP policy decrease after controlling for concurrent policy effects, they remain significantly negative. This indicates that while the policy effects might have been previously overestimated, the mitigating effect of the ZWCP policy on energy consumption intensity remains significant.

4.2.3. Resetting the DML Model

To ensure that our conclusions are not affected by specification bias in double machine learning, we verified the robustness of our baseline regression results based on several aspects. (1) Adjusting sample splitting ratios: While a larger number of K-folds provides more training data, it may increase computational costs and variance; conversely, fewer K-folds might lead to insufficient training samples, affecting prediction accuracy. Therefore, we adjusted the sample splitting ratio from 1:4 in the baseline regression to 1:2 and 1:7 to examine how different splitting ratios affect our findings. (2) Changing machine learning algorithms: Different machine learning algorithms have distinct advantages in handling nonlinear relationships, high-dimensional features, and potential interaction terms. We sequentially replaced the random forest algorithm in the baseline regression with LASSO regression, gradient boosting, and neural networks to explore how different machine learning algorithms influence our conclusions. (3) Modifying model specifications: Although we constructed a partially linear model based on double machine learning algorithms in the baseline analysis, this model specification still maintains some subjectivity. Therefore, to further test the robustness of double machine learning estimates, we introduced more generalizable interactive models for regression analysis to verify our findings under more flexible specifications.

Columns (1) to (6) in

Table 5. Robustness check II.

	(1)	(2)	(3)	(4)	(5)	(6)
Var	Kfolds = 3	Kfolds = 8	Lassocv	Gradboost	Nnet	Interactive Model
Event	−0.035 **	−0.035 **	−0.056 ***	−0.060 ***	−0.093 **	−0.209 ***
	(0.014)	(0.014)	(0.017)	(0.017)	(0.045)	(0.022)
Control (single term)	Y	Y	Y	Y	Y	Y
Control (quadratic term)	Y	Y	Y	Y	Y	Y
City FE	Y	Y	Y	Y	Y	Y
Year FE	Y	Y	Y	Y	Y	Y
N	3230	3230	3230	3230	3230	3230

Note: The dependent variable is energy consumption intensity. Columns (1) and (2) vary the sample split ratio. Columns (3) to (5) employ Lasso regression, gradient boosting, and neural network methods for prediction, respectively. Column (6) adopts an interaction model. Robust standard errors are reported in parentheses. *** p < 0.01; ** p < 0.05.

present the regression results after reconfiguring the double machine learning model. Evidently, neither the sample splitting ratio and machine learning algorithm nor model specification changes alter our conclusion that the ZWCP policy reduces urban energy consumption intensity, only varying the magnitude of the policy effect to some extent. This further demonstrates the robustness of our baseline regression results.

4.2.4. Shortening the Sample Window Period

The choice of time window is crucial for estimation results in double machine learning. A shorter window may lead to insufficient training samples, affecting the prediction accuracy and model convergence of machine learning algorithms, while a longer window might introduce more time-varying confounding factors and increase prediction bias. To verify the robustness of the results, this study narrowed the sample window to periods immediately before and after policy implementation, re-estimating with subsamples from 2013 to 2022 and 2016 to 2022, respectively. As shown in Columns (1) and (2) of

Table 6. Robustness check III.

	(1)	(2)	(3)	(4)
Var	2013–2022	2016–2022	5%	IV
Event	−0.051 ***	−0.064 ***	−0.033 **	−2.349 ***
	(0.013)	(0.012)	(0.014)	(0.822)
Control (single term)	Y	Y	Y	Y
Control (quadratic term)	Y	Y	Y	Y
City FE	Y	Y	Y	Y
Year FE	Y	Y	Y	Y
N	3230	3230	3230	3230

Notes: The dependent variable is energy consumption intensity. Columns (1) and (2) analyze the effects of different policy implementation timings. Column (3) presents the results after 5% winsorization on both tails. Column (4) reports the instrumental variable estimation results. Robust standard errors are reported in parentheses. *** p < 0.01; ** p < 0.05.

, the policy effect coefficients remain significantly negative across different time windows, with the regression coefficients of the ZWCP policy showing an upward trend, strongly demonstrating the robustness of the research findings.

4.2.5. Elimination of Outliers

Different levels of winsorization for outliers in the research sample may lead to estimation bias. Therefore, we winsorized all variables except for the policy variable at the 5th and 95th percentiles, replacing values beyond these thresholds, and re-estimated our baseline regression. The results are presented in Column (3) of Table 6. The estimated coefficient of the ZWCP policy remains significantly negative, confirming the robustness of our baseline findings.

4.2.6. Instrumental Variable

The selection of ZWCP areas is potentially non-random and may be influenced by resource endowment factors, such as the level of economic development, degree of digitalization, and industrial structure, leading to endogeneity concerns. Following Bach et al. [35], we address this issue by constructing a partially linear instrumental variable model with double machine learning:

$${ECI}_{it}={\theta }_0{Event}_{it}+g\left(X_{it}\right)+U_{it}$$

(11)

$${IV}_{it}=m\left(X_{it}\right)+V_{it}$$

(12)

Building upon and improving the approach of Bi et al. [36], we employed the interaction term between regional average slope and river density as an instrumental variable for the ZWCP policy. The validity of this instrument can be justified from both relevance and exclusion perspectives. Regarding relevance, this instrumental variable affects policy implementation through two channels. On the one hand, the construction of zero-waste cities requires enhanced waste treatment efficiency because terrain slope is positively correlated with waste treatment costs. Studies have shown that steep slopes require special transportation equipment such as cable cars, mountainous terrain demands additional slope stability engineering, waste collection vehicles in hilly areas consume more fuel, and the leachate treatment costs of mountain landfills are significantly higher than those in flat areas. All these factors are closely related to the efficiency of zero-waste city construction. On the other hand, river density influences policy implementation by affecting urban solid waste pollution levels. Cities with higher river density typically enjoy more convenient transportation conditions, and this locational advantage tends to attract industrial investment, potentially leading to more severe solid waste pollution problems, which in turn prompts local governments to strengthen their solid waste management policies. For the exclusion restriction, both regional average slope and river density are natural geographical features that serve as exogenous indicators and have no direct correlation with energy consumption intensity. Therefore, the interaction term of these two indicators satisfies both the relevance and exclusion restriction requirements for instrumental variables in econometrics. As reported in Column (4) of Table 6, the estimated coefficient of the policy variable maintains its statistical significance at the 1% level with a negative sign. This aligns with our baseline regression findings.

4.2.7. Causal Forests

Although random forests can handle various features, they are primarily utilized for predictive tasks. While they can be applied to causal inference, this is not their main purpose. In causal inference, random forests may not effectively control for confounding factors, leading to biased estimates of causal effects [37]. In contrast, causal forests are specifically designed for causal inference and can effectively estimate treatment effects. By considering the differences between treatment and control groups, they provide more accurate estimates of causal effects [38]. To this end, we employed the “Causal Forests” command from the R package 4.4.3 “grf”, which provides causal forest models, assessing the robustness of the model with different numbers of trees (500, 2000, 4000, and 8000), while using default settings for parameters such as the number of leaf nodes, minimum sample size, and minimum error. The results are shown in Figure 4.

From the figure, it can be observed that as the number of trees increases, the distribution of the conditional average treatment effect gradually stabilizes. With 500 trees, the distribution of treatment effects is relatively dispersed and exhibits some skewness; however, as the number of trees increases to 2000, 4000, and 8000, the distribution of treatment effects becomes more concentrated, with a negative mean, indicating that the ZWCP policy has a significant effect on reducing urban energy consumption intensity. This result suggests that the estimates from the causal forest model are robust across different numbers of trees, further enhancing our confidence in the policy’s effectiveness.

5. Heterogeneity Test

5.1. Agglomeration Type

The implementation effects of the ZWCP policy may exhibit significant differences due to urban hierarchy and developmental characteristics between cities, which is reflected in the policy execution outcomes between urban and non-urban agglomerations. Based on this, this study systematically reviews relevant policy documents of national-level urban agglomerations and refines the research sample into two categories: urban agglomerations and non-urban agglomerations. Specifically, the urban agglomerations covered in this study include Beijing–Tianjin–Hebei, the Yangtze River Delta, the Pearl River Delta, Chengdu–Chongqing, Guanzhong, Central Plains, Shandong Peninsula, Beibu Gulf, and the coastal urban agglomerations of Guangdong, Fujian, and Zhejiang.

The results of the regression analysis (as shown in Columns (1) and (2) of

Table 7. Heterogeneity test results.

	(1)	(2)	(3)	(4)	(5)	(6)
Var	Urban Cluster	Non-Urban Cluster	Coastal	Inland	Large	Small to Medium
Event	0.001	−0.052 ***	−0.007	−0.080 ***	−0.020	−0.051 **
	(0.017)	(0.019)	(0.016)	(0.024)	(0.023)	(0.022)
Control (single term)	Y	Y	Y	Y	Y	Y
Control (quadratic term)	Y	Y	Y	Y	Y	Y
City FE	Y	Y	Y	Y	Y	Y
Year FE	Y	Y	Y	Y	Y	Y
N	3230	3230	3230	3230	3230	3230

Note: The dependent variable is energy consumption intensity. Robust standard errors are reported in parentheses. *** p < 0.01; ** p < 0.05.

) reveal a complex policy execution mechanism: within urban agglomerations, the ZWCP policy does not significantly suppress urban energy consumption intensity. This phenomenon may stem from the economic structure and developmental inertia within urban agglomerations. Generally, urban agglomerations exhibit a high degree of economic interconnectedness, characterized by complex industrial chains, dense factor flows, and deep regional collaboration. This integrated development model may weaken the direct regulatory effects of any single policy, making it challenging for the ZWCP policy to achieve significant regulatory effects on energy consumption intensity within urban agglomerations [39]. In contrast, for non-urban agglomerations, the ZWCP policy demonstrates a significant suppressive effect on energy consumption intensity. This is primarily due to the relatively limited economic scale and relatively simple industrial structure of cities in non-urban agglomerations, which lack the complex economic networks found in urban agglomerations. In these cities, the ZWCP policy can more directly and effectively influence the urban energy consumption system, optimizing resource allocation and promoting green technology innovation, thereby effectively reducing energy consumption intensity. The research results clearly indicate that the ZWCP policy has a significant suppressive effect on urban energy consumption intensity in non-urban agglomerations, providing important policy support for regional green transformation.

5.2. Geographical Location

Differences in geographical location may lead to significant variations in industrial structure, technological innovation capabilities, and resource endowments among cities, which can affect the energy-saving effects of ZWCP policies. Based on the cities’ geographical locations, this study divides the sample cities into coastal and inland categories to explore the heterogeneous impacts of pilot policies across different regions.

As shown in Columns (3) and (4) of Table 7, the ZWCP policy has a significant inhibitory effect on energy consumption intensity in inland cities, while its impact on coastal cities is not statistically significant. This result can be attributed to the following mechanisms: First, the industrial structure in inland areas is relatively homogeneous, mainly characterized by traditional high-energy-consumption and high-pollution industries. These regions require policy guidance to achieve industrial transformation and upgrading. The ZWCP policy can effectively reduce energy consumption intensity in inland cities by promoting green technological innovation and optimizing industrial structure. Second, compared to developed coastal regions, inland areas have a relatively weak foundation in green innovation and environmental governance. The pilot policy provides crucial institutional incentives for these regions, motivating local governments and enterprises to place greater emphasis on energy conservation and emission reduction. By encouraging green technological innovation and increasing fiscal expenditure on environmental governance, inland cities can better leverage policy dividends and drive the transformation of energy consumption patterns. In contrast, coastal regions already have a relatively sophisticated industrial structure, with higher levels of economic development and technological innovation capabilities. Consequently, the marginal effect of ZWCP policies is relatively weak. Enterprises in these areas may already possess a high awareness of energy conservation and emission reduction, as well as advanced technological capabilities, making the policy’s guiding role less significant compared to inland regions.

5.3. City Scale

The heterogeneity of urban scale may lead to significant variations in the effectiveness of ZWCP policies in managing energy consumption intensity. Based on population scale, this study divides sample cities into large cities and small-to-medium-sized cities to investigate the differentiated impacts of pilot policies across different urban scales.

As shown in Columns (5) and (6) of Table 7, the ZWCP policy demonstrates a significant suppressive effect on energy consumption intensity in small-to-medium-sized cities, while its impact on large cities remains relatively weak. From a mechanism perspective, this phenomenon can be interpreted through multiple dimensions: Large cities generally face more complex energy dependency lock-in challenges. Due to their substantial economic scale and complex industrial structures, the energy consumption systems of large cities exhibit persistent path dependencies, along with high transformation costs that notably constrain the marginal effects of pilot policies. In contrast, small-to-medium-sized cities have relatively simpler industrial structures, clearer transformation pathways, and stronger system resilience and adaptability, enabling them to more agilely respond to policy guidance and rapidly optimize energy consumption patterns.

In terms of innovation dynamics, small-to-medium-sized cities reveal more pronounced potential advantages. These cities typically demonstrate higher policy sensitivity and reform willingness, capable of swiftly and effectively implementing ZWCP policies. These policies provide crucial institutional incentives for small-to-medium-sized cities, effectively reducing energy consumption intensity by encouraging green technological innovation and guiding adjustments in industrial structure. Compared to large cities, small-to-medium-sized cities have more concentrated innovation resources, making it easier to form policy synergies and accelerate the transformation of energy consumption models.

6. Channel Analysis

Based on the aforementioned main findings, it is necessary to further investigate how the ZWCP policy influences urban energy consumption intensity through different mechanisms. Theoretically, the ZWCP policy has three potential mediating mechanisms that explain its impact on urban energy intensity. First, the policy promotes urban industrial structure upgrading. The optimization of industrial structure encourages high-energy-consumption industries to gradually exit or transform into green and low-carbon industries, thereby reducing urban energy consumption intensity. Second, the policy guides energy use toward low carbonization through green technological innovation, promoting the widespread application of energy-saving technologies and significantly improving urban energy utilization efficiency. Finally, policy implementation enhances government and social environmental awareness and, by increasing fiscal expenditure on environmental governance and environmental investment, further strengthens urban energy conservation and emission reduction capabilities. These mechanisms interact, constituting the internal logic of how the ZWCP policy influences urban energy consumption intensity.

6.1. Industrial Structure Transformation

This paper draws on the research methodology of Mishra et al. [40] to construct a comprehensive indicator for industrial structure upgrading. The specific calculation formula is as follows: the proportion of the primary industry in GDP × 1 + the proportion of the secondary industry in GDP × 2 + the proportion of the tertiary industry in GDP × 3. The data are sourced from CNRDS. As shown in Column (1) of

Table 8. Channel analysis.

	(1)	(2)	(3)	(4)
Var	Upgrading	Innovation	Government	Enterprise
Event	0.012 **	0.115 **	0.058 **	0.372 ***
	−0.005	−0.055	−0.024	−0.096
Control (single term)	Y	Y	Y	Y
Control (quadratic term)	Y	Y	Y	Y
City FE	Y	Y	Y	Y
Year FE	Y	Y	Y	Y
N	3230	3230	3230	3230

Notes: The dependent variable is energy consumption intensity. Column (1) tests the mechanism of industrial structure upgrading, Column (2) tests the mechanism of green technology innovation, and Columns (3) and (4) test the mechanism of environmental awareness. Robust standard errors are reported in parentheses. *** p < 0.01; ** p < 0.05.

, the ZWCP policy has a significant positive impact on industrial structure upgrading. This indicates that after the implementation of the policy, pilot cities accelerated their transition to a low-carbon and sustainable industrial structure through policy guidance. The existing literature has confirmed that industrial structure upgrading can effectively alleviate urban energy consumption intensity, primarily achieved through optimizing resource allocation efficiency and promoting industrial collaboration and agglomeration. Specifically, as resources become concentrated in high-efficiency and low-energy-consumption industries, industrial clusters reduce energy usage costs and form a low-carbon, circular economy development model. Therefore, as the regional industrial structure continues to become more rationalized and advanced, the effectiveness of the ZWCP policy in reducing urban energy consumption intensity becomes increasingly prominent.

6.2. Green Technology Innovation

This paper draws on the research methodology of Wang et al. [41] and uses the logarithm of the number of green invention patents applied for in the region to measure the level of green technology innovation. The data are sourced from CNRDS. As shown in Column (2) of Table 8, the ZWCP policy has significantly promoted green technology innovation. After the implementation of the policy, pilot cities encouraged enterprises to increase their investment in green technology research and development through policy guidance, thereby driving the rapid development of green technologies. Specifically, green technology innovation not only enhances resource utilization efficiency but also effectively reduces energy consumption during the production process. With the widespread application of green technologies, enterprises are gradually shifting toward the use of clean energy and environmentally friendly materials, thereby reducing their reliance on traditional high-energy-consumption and high-pollution technologies. Furthermore, the dual drivers of policy support and market demand have facilitated the rapid diffusion of green technologies, creating a favorable innovation ecosystem. In this process, the competitiveness of enterprises has been strengthened, and the economic structure has gradually transitioned toward a low-carbon and sustainable direction. Therefore, the effectiveness of the ZWCP policy in promoting green technology innovation is becoming increasingly evident, providing a solid foundation for achieving urban sustainable development.

6.3. Improvement in Environmental Awareness

The implementation of the ZWCP policy has not only played an important role at the technical and market levels but also had a profound impact on the environmental awareness of governments and enterprises, thereby promoting a reduction in urban energy consumption intensity. The enhancement of environmental awareness can effectively drive collective emission reduction actions across society, encouraging more governments and enterprises to actively participate in energy consumption control, thus reducing the intercity energy consumption inequality caused by differences in environmental awareness. Therefore, this paper comprehensively measures the improvement in environmental awareness from the perspectives of the government and corporate investors. Specifically, environmental awareness at the government level is assessed in reference to the research methodology of Qian et al. [42] and Liu et al. [43], which involves collecting government work reports from the official websites of various municipal governments and extracting 53 key terms from the perspective of environmental governance. Through the counting of the frequency of these keywords and the application of logarithmic transformation, a quantitative indicator of local government environmental awareness is constructed. A higher value of this indicator indicates a greater level of attention paid by local governments to environmental issues. At the corporate level, this paper collects interactive Q&A text between corporate investors and listed companies as raw data and analyzes the content of investor questions. If an investor’s question contains terms related to green and environmental protection, it is thought that the investor is concerned about or values the company’s green and environmental efforts and is recorded as 1; otherwise, it is recorded as 0. Finally, the number of green- and environmental protection-related questions from investors for each listed company in each year is counted and logarithmically transformed to obtain the corporate investor environmental awareness indicator. Figure 5 lists the key words related to the environmental protection concerns of the government and investors. The keywords for the government are located on the outer side of the diagram, while those for investors are on the inner side.

The regression results are shown in Columns (3) and (4) of Table 8. Column (3) indicates that the ZWCP policy has significantly increased the government’s attention on environmental issues. Column (4) shows that corporate investors in the pilot areas exhibit a stronger enthusiasm for reducing energy intensity, which also indirectly suggests that the policy has enhanced corporate environmental responsibility. In summary, the ZWCP policy has improved environmental awareness, prompting society as a whole to participate in energy reduction actions, thereby gradually alleviating urban energy consumption intensity.

7. Conclusions and Policy Recommendations

7.1. Conclusions

The ZWCP policy, as a crucial measure for promoting ecological civilization construction and achieving high-quality development in China, has played a significant role in optimizing the urban energy consumption structure and reducing energy consumption intensity. Through a double machine learning model, this study empirically analyzed the impact of the ZWCP policy on urban energy consumption intensity. The research results indicate that the policy significantly reduced energy consumption intensity in pilot areas. The policy’s impact mechanism is primarily achieved through industrial structure upgrading, green technological innovation, and enhanced environmental awareness. Heterogeneity analysis further revealed that the policy effects are more pronounced in non-urban cluster regions, inland areas, and medium-sized cities, highlighting the significant value of the ZWCP policy in promoting regional green development.

7.2. Policy Recommendations

7.2.1. Deepening the Suppression Mechanism of ZWCP in Energy Consumption Intensity

Deepening the mechanism for suppressing energy consumption intensity requires a comprehensive and systematic approach. Based on the research findings, the policy’s guiding role should be fully leveraged by optimizing industrial structure, promoting technological innovation, and designing institutional frameworks to comprehensively reduce urban energy consumption intensity. Specifically, industrial structure adjustment should be initiated by guiding high-energy-consumption industries to accelerate transformation, focusing on promoting circular development in industrial parks and constructing a resource-circular industrial system to improve resource utilization efficiency. For key industries such as coal, non-ferrous metals, gold, metallurgy, and chemicals, green mining should be implemented according to green mine construction requirements to reduce solid waste generation and storage. Simultaneously, an extended producer responsibility system should be established that requires enterprises to be responsible throughout the product lifecycle, while encouraging the selection of low-environmental-impact products at the source. In terms of fiscal and tax support, the government should encourage waste classification and processing technology development through reward policies and tax exemptions. Specific measures include supporting new voluntary clean production industrial enterprises, green design products, green supply chains, and green industrial design promotion projects and setting indicators for pre-fabricated building proportions and green construction demonstration projects in the construction sector. Technological innovation, as a key mechanism for suppressing energy consumption intensity, requires national and local government funding to create an innovation ecosystem for collaborative solid waste treatment research, encouraging high-tech enterprises to establish technology demonstration and achievement transformation bases. Government guidance should extend to market demand, prioritizing the use of products that comprehensively utilize bulk industrial solid waste through government procurement to guide industrial upgrading. Additionally, a resource- and energy-cascading circular utilization system should be constructed across industrial, agricultural, and living domains, promoting circular economic industrial chain operations within and between enterprises and regions. Moreover, zero-waste city construction should emphasize public participation by transmitting green lifestyle concepts through community activities and science popularization facilities, imbuing physical spaces with a sustainable cultural atmosphere, raising public environmental awareness, and forming a positive societal environment for energy conservation and emission reduction. Through these systematic measures, a virtuous green, low-carbon, circular development ecosystem can be formed, effectively suppressing urban energy consumption intensity and providing strong support for achieving carbon peak and carbon neutrality goals.

7.2.2. Promoting Zero-Waste City Technological Innovation and Ecological Governance

Strengthening the intermediary-effect pathways and promoting technological innovation and ecological governance are core supports for zero-waste city construction. Based on research indicating that industrial structure upgrading, green technological innovation, and enhanced environmental awareness are key mechanisms influencing energy consumption intensity, a comprehensive innovation ecosystem should be constructed. Specifically, by increasing support for green low-carbon development funds, the government will establish an infrastructure system of “industrial compounds, urban–industrial integration, and functional aggregation”, creating comprehensive technology demonstration bases for municipal waste, construction waste, hazardous waste, etc. The focus will be on supporting technological innovations for solid waste source reduction and resource recycling, encouraging collaborative research among enterprises, universities, and research institutions. Simultaneously, improving talent-introduction incentive policies in the solid waste field is crucial for building high-level scientific and technological talent teams. Through the establishment of special funds or research projects, not only can green production and circular utilization be encouraged, but innovation technology market application can be provided with trial opportunities to reduce innovation risks. In the technological innovation pathway, key focus areas include research into and the development of solid waste treatment technologies, reutilization technologies, and environmental pollution control technologies. Efforts should intensify the management of solid waste-processing goods, packaging, and electronic waste, establishing an informative supervision platform for classification, collection, transportation, and disposal. Cultivating environmental governance market entities is the core of the innovation ecosystem. With ecological environment quality improvement as the core and green environmental protection industry expansion as the goal, the government should fully leverage market resource allocation, guiding enterprises to improve environmental public service efficiency and form a diverse environmental governance system. In the transformation of achievements, it is essential to establish an open, collaborative, and shared innovation platform that creates a market trading platform for solid waste, ensuring accurate matching between solid waste generators and processors. Through the establishment of research projects to identify industry needs, achievement transformation can be promoted and collaborative exchange mechanisms established. Importantly, solid waste research platforms can serve public service roles as well, thereby enhancing public engagement and facilitating collaborative governance. By setting up a zero-waste special zone on official websites, sharing construction experiences, utilizing new media platforms for publicity, education, and social supervision, and leveraging collective wisdom to continuously optimize policy implementation methods, these systematic measures can effectively stimulate societal innovation vitality and continuously enhance green technological innovation capabilities, providing robust technological support for zero-waste city construction.

7.2.3. Forming Precise Regional Paths for Zero-Waste City Construction

Addressing the regional characteristics of heterogeneity in zero-waste city construction requires the development of precise policies from multiple, differentiated perspectives. For non-urban cluster areas, the focus should be on supporting the establishment of cross-administrative resource recycling industrial chains, breaking administrative barriers and promoting regional collaborative development. Specifically, this can be achieved by implementing qualification mutual recognition in areas like automobile dismantling and hazardous waste processing, constructing policy and system collaborative mechanisms to break traditional administrative division limitations and promote efficient circulation and the comprehensive utilization of solid waste resources across regions. For inland areas and medium-sized cities, more targeted support strategies are needed. By establishing special funds to encourage green production and circular utilization, lowering market entry barriers, providing targeted technical guidance, and focusing on supporting solid waste processing, reutilization, and environmental pollution control technology research and development, these regions can accelerate green transformation and narrow development gaps with more advanced regions. Simultaneously, localities should be encouraged to propose “Zero-waste Cell” construction forms and pilot scopes in national spatial planning based on their resource endowments and industrial characteristics, exploring locally distinctive zero-waste city construction paths. In institutional innovation, policy integration should be strengthened to enhance the systematicity of pilot schemes. Through the integration of existing circular economy, clean production, and resource utilization reform pilot experiences around zero-waste city construction goals, more targeted implementation plans can be developed through inheritance and innovation. Through the construction of multi-level environmental emergency systems, regional environmental governance capabilities can be improved, promoting a collaborative and efficient governance pattern with top-down and bottom-up linkages. Importantly, zero-waste city construction is not merely technological innovation and management reform but a social transformation. Through the organization of community-level activities like garden creation and public welfare events, green development concepts can be transmitted to the public, inspiring broad social participation and forming an intergenerational ecological civilization consensus. This bottom-up social mobilization will inject continuous innovation vitality into zero-waste city construction. Ultimately, through these differentiated and precise policy supports, regional development gaps can be effectively narrowed and the coordinated development of zero-waste city construction can be promoted. This represents not only an innovation in urban governance models but also an important exploration toward a “Waste-Free Society”, reflecting China’s strategic wisdom and forward-looking vision in ecological civilization construction.

7.3. Limitations and Future Directions

This study also has some limitations and areas for further research. Firstly, the use of DMSP night-time light data as a proxy for actual energy consumption is limited in accuracy, particularly in the absence of empirical validation and the inability to establish effective correlations with official energy statistics. Secondly, our reliance on keyword frequency-based text indicators to measure environmental awareness may be influenced by strategic narratives (such as greenwashing), thus failing to adequately reflect actual behavioral changes. Therefore, future research should consider integrating more quantitative and qualitative methods to provide a more comprehensive assessment of environmental awareness. Recognizing these limitations paves the way for future exploration and encourages further investigation and methodological innovation.