Energy Services Demand Forecasting Combined with Feature Preferences and Bidirectional Long- and Short-Term Memory Networks

doi:10.16183/j.cnki.jsjtu.2023.458

Abstract

Abstract:

Accurate and efficient demand forecasting of customer energy services is crucial for quality and risk management in grid customer service. Therefore, this paper proposes a user energy service demand prediction model based on feature selection. The methodology includes introducing a sampling algorithm to solve the class imbalance problem in the data on the basis of analysing the user energy service data, reducing the dimensionality of the data based on an autoencoder to ensure efficient clustering of the K-mean algorithm, constructing a feature selection algorithm based on a lightweight gradient lifting machine to filter the effective features and improve the training efficiency of the prediction model, and establishing a bidirectional long- and short-term memory neural network multi-label predicting model based on an attentional mechanism to refine the user’s energy service demand. Through the analysis of 720 000 work order data from Guangdong Power Grid over three years, showing that the model proposed can effectively improve the prediction accuracy and speed.

Key words: energy services, demand forecasting, class imbalance, automatic encoder, feature optimization, multi label classification

CLC Number:

TM721

KANG Feng, TAN Huochao, SU Liwei, JIAN Donglin, WANG Shuai, QIN Hao, ZHANG Yongjun. Energy Services Demand Forecasting Combined with Feature Preferences and Bidirectional Long- and Short-Term Memory Networks[J]. Journal of Shanghai Jiao Tong University, 2025, 59(7): 1007-1018.

Figures/Tables 20

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Fig.9

Fig.10

Fig.11

Fig.12

Fig.13

Fig.14

Fig.15

Fig.16

Tab.1

Tab.2

Tab.3

Tab.4

References 22

[1]	BARJA MARTINEZ S. Artificial intelligence techniques for enabling Big Data services in distribution networks: A review[J]. Renewable and Sustainable Energy Reviews, 2021, 150: 111459.
[2]	葛磊蛟, 刘航旭, 孙永辉, 等. 智能配电网多元电力用户群体特性精准感知技术综述[J]. 电力系统自动化, 2023(5): 1-18.
	GE Leijiao, LIU Hangxu, SUN Yonghui, et al. A review of accurate sensing technologies for multiple power user group characteristics in smart distribution networks[J]. Power System Automation, 2023(5): 1-18.
[3]	张勇军, 羿应棋, 李立浧, 等. 双碳目标驱动的新型低压配电系统技术展望[J]. 电力系统自化, 2022, 46(22): 1-12.
	ZHANG Yongjun, YI Yingqi, LI Licheng, et al. Dual-carbon target-driven technology outlook of new low-voltage distribution system[J]. Power System Automation, 2022, 46(22): 1-12.
[4]	张亮, 屈刚, 李慧星, 等. 智能电网电力监控系统网络安全态势感知平台关键技术研究及应用[J]. 上海交通大学学报, 2021, 55 (Sup.2): 103-109. doi: 10.16183/j.cnki.jsjtu.2021.S2.017
	ZHANG Liang, QU Gang, LI Huixing, et al. Research and application of key technology of network security situational awareness platform for smart grid power monitoring system[J]. Journal of Shanghai Jiao Tong University, 2021, 55 (Sup.2): 103-109.
[5]	朱天怡, 艾芊, 贺兴, 等. 基于数据驱动的用电行为分析方法及应用综述[J]. 电网技术, 2020, 40(9):3497-3507.
	ZHU Tianyi, AI Qian, HE Xing, et al. A review of data-driven power usage behaviour analysis methods and applications[J]. Grid Technology, 2020, 40(9): 3497-3507.
[6]	李利娟, 刘海, 刘红良, 等. 融合外部注意力机制的序列到点非侵入式负荷分解[J]. 上海交通大学学报, 2024, 58(6): 846-854. doi: 10.16183/j.cnki.jsjtu.2022.534
	LI Lijuan, LIU Hai, LIU Hongliang, et al. Sequence-to-point non-intrusive load decomposition incorporating external attention mechanisms[J]. Journal of Shanghai Jiao Tong University, 2024, 58(6): 846-854.
[7]	孔祥玉, 马玉莹, 艾芊, 等. 新型电力系统多元用户的用电特征建模与用电负荷预测综述[J]. 电力系统自动化, 2023, 47(13): 2-17.
	KONG Xiangyu, MA Yuying, AI Qian, et al. A review of power usage characteristics modelling and power load forecasting for multiple users in new power systems[J]. Power System Automation, 2023, 47(13): 2-17.
[8]	CAMERO A, LUQUE G, BRAVO Y, et al. Customer segmentation based on the electricity demand signature: The Andalusian case[J]. Energies, 2018, 11(7): 1788-1803.
[9]	朱州. 基于大数据分析的电力客户服务需求预测[J]. 沈阳工业大学学报, 2020, 42(4):368-372. doi: 10.7688/j.issn.1000-1646.2020.04.02
	ZHU Zhou. Electricity customer service demand forecasting based on big data analysis[J]. Journal of Shenyang University of Technology, 2020, 42(4): 368-372. doi: 10.7688/j.issn.1000-1646.2020.04.02
[10]	RAMOS S, SOARES J, CEMBRANEL S S, et al. Data mining techniques for electricity customer characterization[J]. Procedia Computer Science, 2021, 186: 475-488.
[11]	HE Y, WANG M, YU J, et al. Research on the hybrid recommendation method of retail electricity price package based on power user characteristics and multi-attribute utility in China[J]. Energies, 2020, 13(11): 2693-2710.
[12]	王岩, 黄莹, 王文瑾, 等. 电力客户准确定位与立体画像多维构建研究[J]. 武汉理工大学学报(信息与管理工程版), 2023, 45(1): 156-159.
	WANG Yan, HUANG Ying, WANG Wenjin, et al. Research on accurate positioning and multi-dimensional construction of three-dimensional portrait of electric power customers[J]. Journal of Wuhan University of Technology (Information and Management Engineering Edition), 2023, 45(1): 156-159.
[13]	张帝, 王韬, 朱吉然, 等. 基于LSTM-Attention融合的电力客户主动服务推荐方法[J]. 电力科学与技术学报, 2022, 37(2): 213-218.
	ZHANG Di, WANG Tao, ZHU Jiran, et al. A proactive service recommendation method for electric power customers based on LSTM-Attention fusion[J]. Journal of Electric Power Science and Technology, 2022, 37(2): 213-218.
[14]	陈娟, 夏鹏, 梁晓伟, 等. 基于CSPSO-K-means算法的电力客户细分及定制化增值服务系统研究[J]. 微型电脑应用, 2021, 37(10): 90-93.
	CHEN Juan, XIA Peng, LIANG Xiaowei, et al. Research on electricity customer segmentation and customised value-added service system based on CSPSO-K-means algorithm[J]. Microcomputer Applications, 2021, 37(10): 90-93.
[15]	邱桂华, 何引生, 邱楠海, 等. 考虑多气象因子累积影响的光伏发电功率预测[J]. 广东电力, 2022, 35(10): 20-28.
	QIU Guihua, HE Yinsheng, QIU Nanhai, et al. Power prediction of photovoltaic power generation considering the cumulative effects of multiple meteorological factors[J]. Guangdong Electric Power, 2022, 35(10): 20-28.
[16]	李钰, 杨道勇, 刘玲亚, 等. 利用生成对抗网络实现水下图像增强[J]. 上海交通大学学报, 2022, 56(2): 134-142. doi: 10.16183/j.cnki.jsjtu.2021.075
	LI Yu, YANG Daoyong, LIU Lingya, et al. Underwater image enhancement using generative adversarial networks[J]. Journal of Shanghai Jiao Tong University, 2022, 56(2): 134-142.
[17]	龙周, 陈松坤, 王德禹. 基于SMOTE算法的船舶结构可靠性优化设计[J]. 上海交通大学学报, 2019, 53(1): 26-34. doi: 10.16183/j.cnki.jsjtu.2019.01.004
	LONG Zhou, CHEN Songkun, WANG Deyu. Reliability optimisation design of ship structure based on SMOTE algorithm[J]. Journal of Shanghai Jiao Tong University, 2019, 53(1): 26-34.
[18]	汪颖, 杨维, 肖先勇, 等. 基于去噪自编码器网络特征降维与改进小批优化K均值算法的海量用户用电行为聚类及分析[J]. 电力自动化设备, 2022, 42(6): 146-153.
	WANG Ying, YANG Wei, XIAO Xianyong, et al. Clustering and analysis of massive users’ electricity consumption behaviour based on denoising self-encoder network feature dimensionality reduction and improved small batch optimization K-mean algorithm[J]. Power Automation Equipment, 2022, 42(6): 146-153.
[19]	肖庆追, 李捷, 陈鹤峰, 等. 基于组合模型的电力用户用电行为分层分类方法[J]. 电力系统及其自动化学报, 2023, 35(5): 82-88.
	XIAO Qingzhui, LI Jie, CHEN Hefeng, et al. A hierarchical classification method for power users’ electricity consumption behaviour based on combinatorial model[J]. Journal of Power System and Automation, 2023, 35(5): 82-88.
[20]	邵振国, 林潇, 张嫣, 等. 基于特征集重构与多标签分类模型的谐波源定位方法[J]. 电力自动化设备, 2024, 44(2): 147-154.
	SHAO Zhenguo, LIN Xiao, ZHANG Yan, et al. Harmonic source localisation method based on feature set reconstruction and multi-label classification model[J]. Power Automation Equipment, 2024, 44(2): 147-154.
[21]	杨志东, 丁建武, 陈广久, 等. 基于 LightGBM 和LSTM模型的电力大数据异常用电检测方法研究[J]. 电测与仪表, 2023, 21: 1-7.
	YANG Zhidong, DING Jianwu, CHEN Guangjiu, et al. Research on anomalous power usage detection method of electric power big data based on LightGBM and LSTM models[J]. Electrical Measurement and Instrumentation, 2023, 21: 1-7.
[22]	孙欣, 王思敏, 谢敬东, 等. 考虑多维影响因素的改进Transformer-PSO短期电价预测方法[J]. 上海交通大学学报, 2024, 58(9): 1420-1431. doi: 10.16183/j.cnki.jsjtu.2023.065
	SUN Xin, WANG Simin, XIE Jingdong, et al. Improved Transformer-PSO short-term electricity price forecasting method considering multidimensional influencing factors[J]. Journal of Shanghai Jiao Tong University, 2024, 58(9): 1420-1431.

序号	层级	数据来源	类型	内容	采样时间
1	区域	营销管理系统	用户区域	抄表区段	2019-01—2022-01
2	行业	营销管理系统	用户类型	用户电表类型(工/商/居)	2019-01—2022-01
3	用户	营销管理系统	用户信息	行业类别、用户编号、电表数	2019-01—2022-01
4	用户	营销管理系统	用户信息	用户类别、行业类别	2019-01—2022-01
5	用户	营销管理系统	工单信息	业务子类、时间	2019-01—2022-01
6	用户	智慧营业厅	工单信息	业务子类、时间	2019-01—2022-01
5	电表	营销管理系统	电表信息	报装容量	2019-01—2022-01
6	电表	计量系统	电量电费	用户月用电量、电费	2019-01—2022-01

参数	取值
轮数	300
隐藏层	256
学习率	0.02
层数	6

算法	准确率	汉明损失	时间/s
LightGBM(未特征优选)	0.764 9	0.187 8	249.43
LightGBM(特征优选)	0.899 3	0.100 7	126.32
AM-BiLSTM(未特征优选)	0.879 6	0.102 4	227.15
AM-BiLSTM(特征优选)	0.953 2	0.051 1	128.95

算法	准确率	汉明损失	时间/s
LightGBM	0.899 3	0.100 7	126.32
AM-BiLSTM	0.953 2	0.051 1	128.95
CNN-GRU	0.949 5	0.051 7	289.45
BiLSTM	0.833 4	0.157 9	196.85
AM-LSTM	0.871 2	0.193 6	243.15
AM-CNN-LSTM	0.949 8	0.051 2	279.67