地理学报 ›› 2019, Vol. 74 ›› Issue (12): 2592-2603.doi: 10.11821/dlxb201912012

• 资源环境与可持续发展 • 上一篇    下一篇

中国碳强度关键影响因子的机器学习识别及其演进

刘卫东1,2, 唐志鹏1,2(), 夏炎3, 韩梦瑶1, 姜宛贝1   

  1. 1. 中国科学院地理科学与资源研究所 中国科学院区域可持续发展分析与模拟重点实验室,北京 100101
    2. 中国科学院大学资源与环境学院,北京 100049
    3. 中国科学院科技战略咨询研究院,北京 100190
  • 收稿日期:2018-05-08 修回日期:2019-10-17 出版日期:2019-12-25 发布日期:2019-12-25
  • 通讯作者: 唐志鹏
  • 作者简介:刘卫东(1967-), 男, 河北隆化人, 研究员, 中国地理学会会员(S110001202M), 主要从事经济地理和区域发展研究。E-mail:liuwd@igsnrr.ac.cn
  • 基金资助:
    国家重点基础研究发展计划(2016YFA0602804);国家自然科学基金项目(41771135)

Identifying the key factors influencing Chinese carbon intensity using machine learning, the random forest algorithm, and evolutionary analysis

LIU Weidong1,2, TANG Zhipeng1,2(), XIA Yan3, HAN Mengyao1, JIANG Wanbei1   

  1. 1. Key Laboratory of Regional Sustainable Development Modeling, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
    2. College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 100049, China
    3. Institute of Science and Development, CAS, Beijing 100190, China
  • Received:2018-05-08 Revised:2019-10-17 Online:2019-12-25 Published:2019-12-25
  • Contact: TANG Zhipeng
  • Supported by:
    The National Key Research and Development Program of China(2016YFA0602804);National Natural Science Foundation of China(41771135)

摘要:

碳强度影响因子数量众多,通过在众多因子中评估其重要性以识别出关键影响因子进而解析碳强度关键因子的变化规律,是中国2030年碳强度能否实现比2005年下降60%~65%目标的科学基础。传统的回归分析方法对于评估众多因子的重要性存在多重共线性等问题,而机器学习处理海量数据则具有较好的稳健性等优点。本文从能源结构、产业结构、技术进步和居民消费等方面选取了56个中国碳强度影响因子指标,采用随机森林算法基于信息熵评估了1980-2014年逐年各项因子的重要性,通过指标数量与信息熵的对应关系统一筛选出每年重要性最大的前22个指标作为相应年度关键影响因子,最终依据关键影响因子的变化趋势划分了3个阶段作了演进分析。结果发现:1980-1991年,碳强度的关键因子主要以高耗能产业规模及占比、化石能源占比和技术进步为主;1992-2007年,中国经济进入快车道增长时期,服务业占比和化石能源价格对碳强度的影响作用开始显现,居民传统消费的影响作用在增大;2008年全球金融危机后,中国进入经济结构深化调整时期,节能减排力度大大增强,新能源占比和居民新兴消费的影响作用迅速显现。为实现2030年碳强度下降60%~65%目标,优化能源结构和产业结构,促进技术进步,提倡绿色消费,强化政策调控是未来需要采取的主要措施。

关键词: 机器学习, 随机森林, 碳强度, 关键影响因子, 中国

Abstract:

As the Chinese government ratified the Paris Climate Agreement in 2016, the goal of reducing carbon dioxide emissions per unit of gross domestic product (carbon intensity) from 60% to 65% of 2005 levels must now be achieved by 2030. However, as numerous factors influence Chinese carbon intensity, it is key to assess their relative importance in order to determine which are most important. As traditional methods are inadequate for identifying key factors from a range acting simultaneously, machine learning is applied in this research. The random forest (RF) algorithm based on decision tree theory was proposed by Breiman (2001); this algorithm is one of the most appropriate because it is insensitive to multicollinearity, robust to missing and unbalanced data, and provides reasonable predictive results. We therefore identified the key factors influencing Chinese carbon intensity using the RF algorithm and analyzed their evolution between 1980 and 2014. The results of this analysis reveal that dominant factors include the scale and proportion of energy-intensive industries as well as fossil energy proportion and technical progress between 1980 and 1991. As the Chinese economy developed rapidly between 1992 and 2007, effects on carbon intensity were enhanced by service industry proportion and the fossil fuel price such that the influence of traditional residential consumption also increased. The Chinese economy then entered a period of deep structural adjustment subsequent to the 2008 global financial crisis; energy-saving emission reductions were greatly enhanced over this period and effects on carbon intensity were also rapidly boosted by the increasing availability of new energy and its residential consumption. Optimization of energy and industrial structures, promotion of technical progress, green consumption, and the reduction and management of emissions will be key to cutting future carbon intensity levels within China. These approaches will all help to achieve the 2030 goal of reducing carbon emission intensity from 60% to 65% of 2005 levels.

Key words: machine learning, random forest, carbon intensity, key factor, China