基于MARS及其组合模型的安徽省碳达峰的预测研究  

Carbon Peak Prediction in Anhui Province Based on MARS and Several Combination Models

作　　者：胡学平 高文祥 陈书琴 HU Xueping;GAO Wenxiang;CHEN Shuqin(College of Mathematics and Physics,Anqing Normal University,Anqing 246133,China;College of Resource and Environment,Anqing Normal University,Anqing 246133,China)

机构地区：[1]安庆师范大学数理学院,安徽安庆246133 [2]安庆师范大学资源环境学院,安徽安庆246133

出　　处：《环境科学与技术》2024年第10期229-236,共8页Environmental Science & Technology

基　　金：安徽省高校杰出青年科研项目(2023AH020037);安徽省皖江流域种群生态模拟与控制国际联合研究中心;省级研究生教育教学改革研究重点项目(2022jyjxggyj321);省级研究生创新创业竞赛项目(2022cxcyjs027)。

摘　　要：该文基于安徽省1991-2022年的碳排放量数据,研究安徽省碳排放影响因素和预测碳达峰的时间。结果显示,在多种单一机器学习模型中,多元自适应回归样条(MARS)模型的拟合效果最佳,在测试集上的拟合效果较优,且具有较好的鲁棒性,影响安徽省碳排放量的因素重要性排名为:单位GDP能耗>人口城镇化率>人均GDP>农业生产技术>二产比重>人口总数,即经济和技术因素是影响安徽省碳排放量的重要因素;采用加权平均方法(WA)和多元线性回归组合方法(Regression)进一步提高拟合精度,发现Re⁃gression组合方法精度高于WA法和单一MARS模型法;采用情景分析法,设置基准模式、粗放模式和低碳模式预测安徽省碳排放量,结果显示,在基准模式和粗放模式下安徽省碳排放量仍然呈现增加趋势,而在低碳模式下可以在2030年实现碳达峰。为促进安徽省能源转型和经济高质量发展,推动中国“双碳”目标的实现,该文提出了转变经济发展模式,提升人均GDP、加快技术进步特别是农业生产技术的发展,降低能源强度,优化能源消费结构等对策。Based on the carbon emission data of Anhui Province from 1991 to 2022,we explore the influencing factors of carbon emissions and predict the time of carbon peak in Anhui Province,which is of great significant to the energy transformation and high-quality economic development and contribution to wisdom and strength of Anhui in the realization of the national"double carbon"goals.Firstly,it is found that the multiple adaptive regression spline(MARS)model has the best fitting effect by comparing the results of multiple single machine learning models fitting effects,and the optimized hyper-parameter model performs well on the test set and has good robustness,the analysis indicates that the importance order of the influencing factors carbon emissions in Anhui Province is as energy consumption per unit GDP>population urbanization rate>per capita GDP>agricultural production technology>proportion of secondary industry>total population,that is,economic and techno⁃logical factors are important influencing factors of carbon emissions.Secondly,the weighted average method(WA)and multiple linear regression combination method(Regression)in the combination model were used to further improve the fitting accuracy,and the results show that the regression combination method has higher accuracy than the WA method and the single MARS method.Finally,scenario analysis is used to predict the carbon emissions by setting baseline mode,extensive mode,and low-carbon mode.It is found that carbon emissions are still showing an increasing trend in both baseline and extensive modes,while carbon peak can be achieved before 2030 in the low-carbon mode.According to this,some emission reduction paths are proposed,including transforming the economic development model and increasing per capita GDP;accelerating tech⁃nological progress,especially the development of agricultural production technology,reducing energy intensity and optimizing energy consumption structure.

关 键 词：碳达峰 低碳模式 MARS模型 组合预测法 

分 类 号：X321[环境科学与工程—环境工程] F127[经济管理—世界经济]