基于空间相似度和深度学习的中长期用电量预测Medium and Long-term Electricity Consumption Prediction Based on Spatial Similarity and Deep Learning
章剑光,刘理峰,林海峰,张永建
ZHANG Jianguang,LIU Lifeng,LIN Haifeng,ZHANG Yongjian
摘要(Abstract)：
准确地预测用户中长期用电量对电力系统优化与调度具有重要意义。为此,提出一种根据用电曲线空间相似性划分用电用户类别,进而利用深度网络预测单类用户中长期用电量的方法,用以提升预测精度。首先,计算用电数据间的动态时间规整距离,基于规整距离利用层次聚类法绘制层次聚类树对用户分类。然后,通过离差标准化约束分类后每类用户数据的取值范围,进而通过深度神经网络建立单类用户的中长期用电量预测模型。最后,通过实例分析了传统方法与所提方法的用户聚类效果,并对比单类用户的总体、个体用电量预测结果,证明了所提依据空间形状相似度指标可较准确地划分用户类别,提升了中长期用电量的预测精度。
Accurate prediction of medium and long-term electricity consumption is of significance to power system optimization and dispatch. This paper proposes a method to classify the electricity users according to the spatial similarity of the electricity consumption curve, and then to predict the medium and long-term electricity consumption of single users by using the deep network to improve the prediction accuracy. Firstly,the dynamic time structuring distance between electricity consumption data is calculated. Based on the structured distance, a hierarchical clustering tree is drawn to classify users. Then, the value range of each class of user data is classified by the deviation standardized constraint, and the medium and long-term electricity consumption prediction model of single users is established through the deep neural network. Finally, the clustering effect of the traditional method and the proposed method is analyzed via an example. The overall and individual electricity consumption prediction results of single-category users are compared. It is proved that the proposed spatial shape similarity index can accurately classify the user categories and improve the prediction accuracy of medium and long-term electricity consumption.
关键词(KeyWords)：
用户聚类;动态时间规整;深度网络;用电量预测
users clustering;dynamic time structuring;deep network;power consumption prediction
基金项目(Foundation): 国家电网有限公司科技项目（5211SX1800AX）
作者(Author):
章剑光,刘理峰,林海峰,张永建
ZHANG Jianguang,LIU Lifeng,LIN Haifeng,ZHANG Yongjian
DOI: 10.19585/j.zjdl.202105007
参考文献(References)：
- [1]张素香，刘建明，赵丙镇，等.基于云计算的居民用电行为分析模型研究[J].电网技术，2013,37(6):1542-1546.
- [2]袁鸣峰，刘陶，山宪武，等.基于行业聚类的负荷特性分析及预测[J].电力系统及其自动化，2019,41(5):77-79.
- [3]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，2020,44(9):3497-3507.
- [4]CH魪VEZ P,BARBERO D,MARTINI I,et al.Application of the k-means clustering method for the detection and analysis of areas of homogeneous residential electricity consumption at the Great La Plata region,Buenos Aires,Argentina[J].Sustainable Cities and Society,2017,32:115-129.
- [5]刘洋，刘洋，许立雄.适用于海量负荷数据分类的高性能反向传播神经网络算法[J].电力系统自动化，2018,42(21):96-103.
- [6]YANG J,ZHAO J,WEN F,et al.A model of customizing electricity retail prices based on load profile clustering analysis[J].IEEE Transactions on Smart Grid,2018,10(3):3374-3386.
- [7]费丹雄，严思唯，芦金雨，等.基于混合高斯模型的用电量计量数据聚类算法研究[J].电子设计工程，2020,28(20):106-110.
- [8]HAGAN M,BEHR S.The time series approach to short term load forecasting[J].IEEE Transactions on Power Systems,1987,2(3):785-791.
- [9]CHEN C,ZHOU J.Appncation of regression analysis in power system lood forecosting[J].Advanced Materids Reseanch,2014,960:1516-1522.
- [10]KHAN G,KHATTAK A,ZAFARI F,et al.Electrical load forecasting using fast learning recurrent neural networks[J].IEEE Transactions on Power System,2013,38(2):1-6.
- [11]HOCHREITER S,SCHMIDHUBER J.Long short-term memory[J].Neural computation,1997,9(8):1735-1780.
- [12]刘云霞.基于动态时间规整的面板数据聚类方法研究及应用[J].统计研究，2016,33(11):93-101.
- [13]GIORGINO T.Computing and visualizing dynamic time warping alignments in R:The dtw package[J].Journal of Statistical Software,2009,31(7):1-24.