摘要
针对用户级综合能源系统(integratedenergysystem,IES)规模小、负荷波动大、能量耦合复杂的特点,提出一种考虑多能时空耦合的超短期负荷预测方法。首先,结合K-means聚类方法和Pearson相关系数,将无明显规律的各类基本负荷单元进行"像素重构",使之在水平和竖直两个方向具有一定的关联特征;其次,利用多通道卷积神经网络(multi-channel convolutional neural network,MCNN)对多类重构后的二维负荷像素在高维空间进行特征的独立提取和统一融合;最后,将扩展气象与节假日信息的综合特征按照时序的方式输入长短时记忆网络(long short-term memory,LSTM)进行负荷预测。以某用户级IES实测负荷数据为算例进行分析,结合基本卷积神经网络(convolutionalneuralnetwork,CNN),对比是否进行像素重构和负荷特征融合的各场景下LSTM、CNN-LSTM、MCNN-LSTM方法的预测效果,结果表明,考虑像素重构和负荷特征融合的MCNN-LSTM方法可有效提高用户级IES负荷的预测精度。
Given the characteristics of small scale,large load fluctuations and complex energy coupling of user-level integrated energy system(IES),an ultra short-term load forecasting method is presented for multi-energy spatio-temporal coupling.Firstly,all the basic load cells without obvious regularity are"pixel reconstructed"by means of K-means clustering method and Pearson correlation coefficient so that the cells show certain correlation both in horizontal and vertical directions.Then the high-dimensional spatial features of the reconstructed two-dimensional load pixels are extracted independently but fused uniformly through the multi-channel convolutional neural network(MCNN).Finally,the integrated features of expanded weather and holiday information are input into the long short-term memory network(LSTM)based on time sequences for load forecasting.Taking the actual load data of a user-level IES as an example and combining with convolutional neural network(CNN),the prediction results of LSTM,CNN-LSTM and MCNN-LSTM are analyzed and compared in scenarios of whether pixel reconstruction and load feature fusion are performed.Results show that the load prediction accuracy of the user-level IES is effectively improved through MCNN-LSTM considering pixel reconstruction and energy feature fusion.
作者
栗然
孙帆
丁星
韩怡
刘英培
严敬汝
LI Ran;SUN Fan;DING Xing;HAN Yi;LIU Yingpei;YAN Jingru(School of Electrical and Electronic Engineering,North China Electric Power University,Baoding 071003,Hebei Province,China;State Grid Hebei Electric Power Research Institute,Shijiazhuang 050021,Hebei Province,China)
出处
《电网技术》
EI
CSCD
北大核心
2020年第11期4121-4131,共11页
Power System Technology
基金
国家自然科学基金项目(51607069)。
关键词
超短期负荷预测
综合能源系统
多通道卷积神经网络
长短时记忆网络
负荷像素
ultra short-term load forecasting
integrated energy system
multi-channel convolutional neural network
long short-term memory network
load pixel
作者简介
通信作者:孙帆(1995),男,硕士研究生,研究方向为综合能源系统负荷预测与运行优化,E-mail:15932266113@163.com;栗然(1965),女,博士,教授,研究方向为新能源与并网技术、电力系统分析及运行与控制.E-mail:liranlelele@163.com;丁星(1995),女,硕士研究生,研究方向为分布式能源规划与调度;韩恰(1995),女,硕士研究生,研究方向为含新能源的电力系统调度;刘英培(1982),女,副教授,研究方向为直流输电等.E-mail:liuyingpei_123@126com;严敬汝(1993),女,助理工程师,研究方向为电力系统分析、运行与控制。
