<p>With the rapid socioeconomic development at the regional level, the service industry carbon emissions (SICE) in provincial level have exhibited a persistent upward trajectory. There is a pressing need to develop a tailored, industry-specific carbon emission analysis model to support scientifically grounded low-carbon development strategies for the region. Firstly, utilizing the IPCC carbon emission factor approach, the paper calculates the SICE in Sichuan during 2005–2022. Secondly, based on the eXtreme Gradient Boosting (XGBoost) model, we quantify the importance of seven influencing factors, such as POP (population size) and GDP. Subsequently, we construct a two-layer Stacking integrated hybrid machine learning model by integrating Particle Swarm Optimization-Extreme Learning Machine (PSO-ELM), Particle Swarm Optimization-Long Short-Term Memory (PSO-LSTM), Support Vector Regression (SVR), and Multilayer Perceptron (MLP) models. Finally, five scenarios are established—Baseline Scenario (BAS), Energy Efficiency Improvement Scenario (EEI), Efficient Growth Scenario (EGS), Low Carbon Development Scenario (LCD), and Accelerated Urbanization Scenario (AUS)—to simulate future carbon emissions and characterize carbon reduction potential in Sichuan's service industry. Results show that: (1) EC (Energy intensity) and UR are the key drivers of SICE in Sichuan, with the sum of their importance scores reaching 0.737. (2) The Stacking integrated hybrid machine learning model better captures the complementarity among base models and demonstrates good overall predictive capability and detail-capturing ability, with MAE, RMSE, MAPE and <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10668_2025_6996_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">\({R}^{2}\)</EquationSource> </InlineEquation> values of 1.6891, 2.1950, 6.8924% and 0.9357 respectively. (3) Under all five scenarios, the SICE in Sichuan exhibit an inverted U-shaped trend, rising initially before declining. Among these, the EGS represents the optimal development pathway, enabling Sichuan's service industry to achieve its 2030 carbon peak at 47.344 million tons with appropriate environmental costs. The findings of this paper offer meaningful insights for enriching research on low-carbon development in regional service industries.</p>

Analysis of key drivers and scenario simulation for carbon emissions in regional service industries: An integrated hybrid machine learning model—A case study of Sichuan Province, China

期刊论文
发表日期：

2025年11月10日

查看全文

Xin Liu,
Siying Wang,
Tingting Feng,
Yue Zhao,
Jie Liu,
Fangfang Shi,
Ranbo Zhang,
Qiyuan Deng

摘要

With the rapid socioeconomic development at the regional level, the service industry carbon emissions (SICE) in provincial level have exhibited a persistent upward trajectory. There is a pressing need to develop a tailored, industry-specific carbon emission analysis model to support scientifically grounded low-carbon development strategies for the region. Firstly, utilizing the IPCC carbon emission factor approach, the paper calculates the SICE in Sichuan during 2005–2022. Secondly, based on the eXtreme Gradient Boosting (XGBoost) model, we quantify the importance of seven influencing factors, such as POP (population size) and GDP. Subsequently, we construct a two-layer Stacking integrated hybrid machine learning model by integrating Particle Swarm Optimization-Extreme Learning Machine (PSO-ELM), Particle Swarm Optimization-Long Short-Term Memory (PSO-LSTM), Support Vector Regression (SVR), and Multilayer Perceptron (MLP) models. Finally, five scenarios are established—Baseline Scenario (BAS), Energy Efficiency Improvement Scenario (EEI), Efficient Growth Scenario (EGS), Low Carbon Development Scenario (LCD), and Accelerated Urbanization Scenario (AUS)—to simulate future carbon emissions and characterize carbon reduction potential in Sichuan's service industry. Results show that: (1) EC (Energy intensity) and UR are the key drivers of SICE in Sichuan, with the sum of their importance scores reaching 0.737. (2) The Stacking integrated hybrid machine learning model better captures the complementarity among base models and demonstrates good overall predictive capability and detail-capturing ability, with MAE, RMSE, MAPE and \({R}^{2}\) values of 1.6891, 2.1950, 6.8924% and 0.9357 respectively. (3) Under all five scenarios, the SICE in Sichuan exhibit an inverted U-shaped trend, rising initially before declining. Among these, the EGS represents the optimal development pathway, enabling Sichuan's service industry to achieve its 2030 carbon peak at 47.344 million tons with appropriate environmental costs. The findings of this paper offer meaningful insights for enriching research on low-carbon development in regional service industries.