基于卷积神经网络与轻量级梯度提升树组合模型的电力行业短期以电折碳方法

doi:10.16183/j.cnki.jsjtu.2023.382

摘要/Abstract

摘要：

电力行业是碳排放的重点控排行业,准确、实时的电力行业碳排放计量是支撑其降碳减排的基础.目前,电力行业的碳排放计量主要基于实测法或核算法,难以很好地兼顾低计量成本与实时计量能力.为此,充分考虑电力行业良好的电力数据基础,挖掘电-碳间的相关关系,以电力历史数据为基础,基于机器学习方法提出一种电力行业短期以电折碳方法,实时估算电力行业短期碳排放情况.该方法使用卷积神经网络进行特征提取,并采用轻量级梯度提升树算法开展基于特征提取值的碳排放测算.此外,为了提升模型的泛化能力和鲁棒性,在模型训练中采用K折交叉验证技术,在模型参数优化过程中采用网格搜索技术.最后,为了验证所提模型的有效性,对比所提模型和其他机器学习模型在同等数据集划分条件下分别基于日度数据集与小时数据集中进行训练的效果.结果表明:所提模型在效果评估和测算值与目标值分布分析中均优于其他模型,能够较好地反映电力行业的短期碳排放情况.

关键词: 以电折碳, 卷积神经网络, 轻量级梯度提升树算法, 碳排放, 机器学习, 组合模型

Abstract:

The electric power industry plays a pivotal role in carbon emission control. Accurate and real-time accounting of carbon emissions in the power industry is essential for supporting the carbon reduction of the power industry. At present, the measurement of carbon emissions in the power industry relies mainly on direct measurement or the accounting methods, which often struggles to balance low measurement costs with real-time accuracy. Therefore, in this paper, the robust power data infrastructure in the power industry is leveraged and the correlation between electricity consumption and carbon emissions is explored to propose a short-term electricity-to-carbon method using machine learning methods based on historical data of electricity. This method utilizes convolutional neural networks (CNNs) for feature extraction, and light gradient boosting machine (LightGBM) for carbon emission estimation based on extracted features. Moreover, K-fold cross-validation is used in model training, with parameter optimization using grid search to enhance the generalization capability and robustness of the model. To validate the proposed method, it is compared with other machine learning models under the same data segmentation condition for daily and hourly data sets. The results indicate that the proposed model outperforms other models in both performance evaluation and the consistency between estimated and target values.

Key words: carbon emission accounting using electricity data, convolutional neural networks (CNNs), light gradient boosting machine (LightGBM), carbon emissions, machine learning, combined model

中图分类号:

TM732

曾金灿, 何耿生, 李姚旺, 杜尔顺, 张宁, 朱浩骏. 基于卷积神经网络与轻量级梯度提升树组合模型的电力行业短期以电折碳方法[J]. 上海交通大学学报, 2025, 59(6): 746-757.

ZENG Jincan, HE Gengsheng, LI Yaowang, DU Ershun, ZHANG Ning, ZHU Haojun. A Short-Term Carbon Emission Accounting Method for Power Industry Using Electricity Data Based on a Combined Model of CNN and LightGBM[J]. Journal of Shanghai Jiao Tong University, 2025, 59(6): 746-757.

图/表 12

图1

表1

图2

表2

图3

图4

图5

图6

表3

图7

表4

图8

参考文献 23

[1]	康重庆, 杜尔顺, 李姚旺, 等. 新型电力系统的“碳视角”: 科学问题与研究框架[J]. 电网技术, 2022, 46(3): 821-833.
	KANG Chongqing, DU Ershun, LI Yaowang, et al. Key scientific problems and research framework for carbon perspective research of new power systems[J]. Power System Technology, 2022, 46(3): 821-833.
[2]	包维瀚, 李姚旺, 季节, 等. 储能系统与双向电力负荷的碳排放核算方法[J]. 电网技术, 2023, 47(8): 3049-3058.
	BAO Weihan, LI Yaowang, JI Jie, et al. Carbon emission accounting method for energy storage system and bidirectional power load[J]. Power System Technology, 2023, 47(8): 3049-3058.
[3]	魏军晓, 耿元波, 王松. 中国水泥碳排放测算的影响因素分析与不确定度计算[J]. 环境科学学报, 2016, 36(11): 4234-4244.
	WEI Junxiao, GENG Yuanbo, WANG Song. Identification of factors influencing CO₂ emission estimation from Chinese cement industry and determination of their uncertainty[J]. Acta Scientiae Circumstantiae, 2016, 36(11): 4234-4244.
[4]	马凯, 韩文涛, 丁艺, 等. 煤种对燃煤电厂碳排放经济性的影响研究[J]. 热能动力工程, 2018, 33(9): 142-146.
	MA Kai, HAN Wentao, DING Yi, et al. Study on the influence of coal on the carbon emission economy of coal-fired power plant[J]. Journal of Engineering for Thermal Energy and Power, 2018, 33(9): 142-146.
[5]	刘学之, 孙鑫, 朱乾坤, 等. 中国二氧化碳排放量相关计量方法研究综述[J]. 生态经济, 2017, 33(11): 21-27.
	LIU Xuezhi, SUN Xin, ZHU Qiankun, et al. Review on the measurement methods of carbon dioxide emissions in China[J]. Ecological Economy, 2017, 33(11): 21-27.
[6]	王安静, 冯宗宪, 孟渤. 中国30省份的碳排放测算以及碳转移研究[J]. 数量经济技术经济研究, 2017, 34(8): 89-104.
	WANG Anjing, FENG Zongxian, MENG Bo. Mea-sure of carbon emissions and carbon transfers in 30 provinces of China[J]. The Journal of Quantitative & Technical Economics, 2017, 34(8): 89-104.
[7]	吴昊, 任鑫, 朱俊杰. 发电行业二氧化碳排放监测技术现状与综述[J]. 热力发电, 2023, 52(7): 1-13.
	WU Hao, REN Xin, ZHU Junjie. Current situation and review of carbon dioxide emission monitoring technology in power generation industry[J]. Thermal Power Generation, 2023, 52(7): 1-13.
[8]	刘昱良, 李姚旺, 周春雷, 等. 电力系统碳排放计量与分析方法综述[J]. 中国电机工程学报, 2024, 44(6): 2220-2236.
	LIU Yuliang, LI Yaowang, ZHOU Chunlei, et al. Overview of carbon measurement and analysis methods in power systems[J]. Proceedings of the CSEE, 2024, 44(6): 2220-2236.
[9]	KANG C Q, ZHOU T R, CHEN Q X, et al. Carbon emission flow from generation to demand: A network-based model[J]. IEEE Transactions on Smart Grid, 2015, 6(5): 2386-2394.
[10]	张宁, 李姚旺, 黄俊辉, 等. 电力系统全环节碳计量方法与碳表系统[J]. 电力系统自动化, 2023, 47(9): 2-12.
	ZHANG Ning, LI Yaowang, HUANG Junhui, et al. Carbon measurement method and carbon meter system for whole chain of power system[J]. Automation of Electric Power Systems, 2023, 47(9): 2-12.
[11]	李姚旺, 刘昱良, 杨晓斌, 等. 计及电量交易信息的用电碳计量方法[J]. 中国电机工程学报, 2024, 44(2): 439-450.
	LI Yaowang, LIU Yuliang, YANG Xiaobin, et al. Electricity carbon metering method considering electricity transaction information[J]. Proceedings of the CSEE, 2024, 44(2): 439-450.
[12]	刘红琴, 王高天, 陈品文, 等. 地区电力行业碳排放水平测算及其特点分析[J]. 生态经济, 2018, 34(4): 34-39.
	LIU Hongqin, WANG Gaotian, CHEN Pinwen, et al. The level measure and characteristics analysis of carbon emission in regional power industry[J]. Ecological Economy, 2018, 34(4): 34-39.
[13]	胡壮丽, 罗毅初, 蔡航. 城市电力行业碳排放测算方法及减碳路径[J]. 上海交通大学学报, 2024, 58(1): 82-90. doi: 10.16183/j.cnki.jsjtu.2022.222
	HU Zhuangli, LUO Yichu, CAI Hang. A method for carbon emission measurement and a carbon reduction path of urban power sector[J]. Journal of Shanghai Jiao Tong University, 2024, 58(1): 82-90.
[14]	李政, 陈思源, 董文娟, 等. 碳约束条件下电力行业低碳转型路径研究[J]. 中国电机工程学报, 2021, 41(12): 3987-4001.
	LI Zheng, CHEN Siyuan, DONG Wenjuan, et al. Low carbon transition pathway of power sector under carbon emission constraints[J]. Proceedings of the CSEE, 2021, 41(12): 3987-4001.
[15]	王丽娟, 张剑, 王雪松, 等. 中国电力行业二氧化碳排放达峰路径研究[J]. 环境科学研究, 2022, 35(2): 329-338.
	WANG Lijuan, ZHANG Jian, WANG Xuesong, et al. Pathway of carbon emission peak in China’s electric power industry[J]. Research of Environmental Sciences, 2022, 35(2): 329-338.
[16]	ZHAO J J, KOU L, WANG H T, et al. Carbon emission prediction model and analysis in the Yellow River Basin based on a machine learning method[J]. Sustainability, 2022, 14(10): 6153.
[17]	LI M L, WANG W, DE G, et al. Forecasting carbon emissions related to energy consumption in Beijing-Tianjin-Hebei Region based on grey prediction theory and extreme learning machine optimized by support vector machine algorithm[J]. Energies, 2018, 11(9): 2475.
[18]	ARAS S, VAN M H. An interpretable forecasting framework for energy consumption and CO₂ emissions[J]. Applied Energy, 2022, 328: 120163.
[19]	徐勇戈, 宋伟雪. 基于FCS-SVM的建筑业碳排放预测研究[J]. 生态经济, 2019, 35(11): 37-41.
	XU Yongge, SONG Weixue. Carbon emission prediction of construction industry based on FCS-SVM[J]. Ecological Economy, 2019, 35(11): 37-41.
[20]	叶鎏芳, 钟志鹏, 郑仁广, 等. 基于碳电强度的碳排放监测方法[J]. 能源与环境, 2023(1): 40-44.
	YE Liufang, ZHONG Zhipeng, ZHENG Renguang, et al. Carbon emission monitoring method based on carbon electric intensity[J]. Energy & Environment, 2023(1): 40-44.
[21]	章琳, 袁非牛, 张文睿, 等. 全卷积神经网络研究综述[J]. 计算机工程与应用, 2020, 56(1): 25-37. doi: 10.3778/j.issn.1002-8331.1910-0164
	ZHANG Lin, YUAN Feiniu, ZHANG Wenrui, et al. Review of fully convolutional neural network[J]. Computer Engineering & Applications, 2020, 56(1): 25-37.
[22]	TORRES J F, HADJOUT D, SEBAA A, et al. Deep learning for time series forecasting: A survey[J]. Big Data, 2021, 9(1): 3-21.
[23]	CAO Q, WU Y H, YANG J, et al. Greenhouse temperature prediction based on time-series features and LightGBM[J]. Applied Sciences, 2023, 13(3): 1610.

数据规模	时间范围	颗粒度
2843×39	2015-07-01-2023-04-12	1 d
68233×39	2015-07-01T01:00:00— 2023-04-13T04:00:00	1 h

时间周期	特征名称	Spearman相关系数	Pearson相关系数
日度	X₁-Y	0.884 9	0.894 8
	X₂-Y	0.849 8	0.888 4
	X₃-Y	0.849 9	0.888 6
	X₄-Y	0.854 1	0.890 8
小时	X₁-Y	0.896 5	0.918 1
	X₂-Y	0.891 8	0.920 5
	X₃-Y	0.892 5	0.920 9
	X₄-Y	0.895 2	0.922 4

K折	模型方法	评估指标
K折	模型方法	e_RMSE	e_MAPE	R²
K=1	Ridge	0.315	6.122	0.852
	CNN	0.316	21.749	0.852
	LightGBM	0.320	6.348	0.847
	CNN-LightGBM	0.293	5.687	0.872
K=2	Ridge	0.331	6.785	0.832
	CNN	0.348	22.810	0.814
	LightGBM	0.337	7.046	0.826
	CNN-LightGBM	0.331	6.857	0.832
K=3	Ridge	0.326	6.503	0.835
	CNN	0.378	24.190	0.778
	LightGBM	0.337	6.914	0.823
	CNN-LightGBM	0.300	5.929	0.860
K=4	Ridge	0.339	6.805	0.830
	CNN	0.346	23.370	0.823
	LightGBM	0.341	7.077	0.828
	CNN-LightGBM	0.310	6.178	0.858
K=5	Ridge	0.335	6.577	0.819
	CNN	0.356	21.831	0.796
	LightGBM	0.340	6.930	0.813
	CNN-LightGBM	0.323	6.555	0.832
K平均	Ridge	0.329	6.559	0.833
	CNN	0.349	22.790	0.812
	LightGBM	0.335	6.863	0.827
	CNN-LightGBM	0.311	6.241	0.851

K折	模型方法	评估指标
K折	模型方法	e_RMSE	e_MAPE	R²
K=1	Ridge	1.480	7.210	0.870
	CNN	1.520	25.190	0.859
	LightGBM	1.550	7.680	0.857
	CNN-LightGBM	1.450	7.060	0.876
K=2	Ridge	1.476	7.189	0.873
	CNN	1.584	28.290	0.854
	LightGBM	1.547	7.712	0.861
	CNN-LightGBM	1.529	7.562	0.864
K=3	Ridge	1.453	7.077	0.879
	CNN	1.483	27.723	0.874
	LightGBM	1.536	7.674	0.864
	CNN-LightGBM	1.414	6.869	0.885
K=4	Ridge	1.550	7.717	0.857
	CNN	1.501	25.435	0.867
	LightGBM	1.551	7.717	0.858
	CNN-LightGBM	1.445	7.036	0.876
K=5	Ridge	1.462	7.136	0.873
	CNN	1.506	27.291	0.866
	LightGBM	1.523	7.592	0.862
	CNN-LightGBM	1.434	7.008	0.878
K平均	Ridge	1.472	7.163	0.873
	CNN	1.519	26.785	0.865
	LightGBM	1.542	7.675	0.861
	CNN-LightGBM	1.455	7.100	0.876