地理科学进展 ›› 2019, Vol. 38 ›› Issue (6): 851-860.doi: 10.18306/dlkxjz.2019.06.006

• 研究论文 • 上一篇    下一篇

北京城市代谢预测研究——基于长短期记忆神经网络模型

刘炳春1(), 齐鑫1, 王庆山2,*()   

  1. 1. 天津理工大学管理学院,天津 300384
    2. 天津农学院,天津 300384
  • 收稿日期:2018-10-04 修回日期:2019-03-18 出版日期:2019-06-28 发布日期:2019-06-27
  • 通讯作者: 王庆山
  • 作者简介:

    第一作者简介:刘炳春(1980— ),男,河北衡水人,副教授,主要研究方向为深度学习与数据挖掘、循环经济。E-mail: tjutlbc@163.com

  • 基金资助:
    国家自然科学基金项目(71503180);天津市教委社会科学重大项目(2017JWZD16)

Urban metabolism prediction of Beijing City based on long short-term memory neural network

Bingchun LIU1(), Xin QI1, Qingshan WANG2,*()   

  1. 1. School of Management, Tianjin University of Technology, Tianjin 300384, China
    2. Tianjin Agricultural University, Tianjin 300384, China
  • Received:2018-10-04 Revised:2019-03-18 Online:2019-06-28 Published:2019-06-27
  • Contact: Qingshan WANG
  • Supported by:
    National Natural Science Foundation of China, No. 71503180;Major Social Science Projects of Tianjin Education Commission, No. 2017JWZD16.

摘要:

城市化进程提升促使城市环境污染加剧、能源消耗激增、人口密度过大等问题的深层次原因在于城市代谢失调。为精准预测北京市城市代谢变化趋势,论文通过能源消费量及人类活动时间指标测算了1980—2016年北京市体外能代谢率,表征城市代谢程度。据此运用长短期记忆神经网络模型(LSTM)预测了2017—2022年北京各部门体外能代谢率。结果表明:① 基于长短期记忆神经网络的城市代谢预测模型精度较高,能够对北京各部门体外能代谢率进行更为精准的预测;② 2017—2022年间,北京第一产业和总体外能代谢率呈下降趋势,其中第一产业在2017年达到峰值,第二、第三产业及生活部门体外能代谢率将呈现增长趋势。③ 除第一、第三产业和总体外能代谢率外,历史变化的时间扰动幅度先小后大。④ 对各部门体外能代谢率EMRT的影响贡献度最大的因子为第二产业体外能代谢率EMR2,最小的为第一产业体外能代谢率EMR1。论文研究结果可为政策制定者优化城市管理方案、提升城市综合实力提供理论依据和决策支持。

关键词: 长短期记忆神经网络, 体外能代谢率, 城市代谢, 北京

Abstract:

The underlying causes of aggravating urban environmental pollution, escalating energy consumption, population overcrowding, and other urban environmental problems are imbalances in urban metabolism. In order to accurately predict the trend of urban metabolism changes in Beijing City, the exosomatic metabolic rate of Beijing from 1980 to 2016 was estimated by the indicators of energy consumption and human activity time, and the degree of urban metabolism was characterized. Based on the results, the long short-term memory (LSTM) neural network model was used to predict the exosomatic metabolic rate of various sectors in Beijing from 2017 to 2022. The results show that: 1) the urban metabolic prediction model based on LSTM neural network has high accuracy and can make more accurate prediction on the exosomatic metabolic rate of various sectors in Beijing. 2) From 2017 to 2022, the exosomatic metabolic rate of the primary industry and the overall external energy in Beijing show a downward trend, among which the primary industry reached its peak in 2017, and the exosomatic metabolism rates of the secondary and tertiary industries show an increasing trend. 3) Except for the primary industry, tertiary industry and the overall exosomatic metabolic rate, the temporal perturbation of historical change ranged from small to large. 4) The factors that contribute the most to EMRT are EMR2, and the least are EMR1. This study may provide a theoretical basis and decision-making support for policymakers to optimize urban management plans and enhance urban comprehensive strength.

Key words: long short-term memory neural network, exosomatic metabolic rate, urban metabolism, Beijing