短时间尺度用电行为相关性分析网络模型

用电行为短时间尺度相关关系是用户精细化管理与智能配用电的重要决策基础。建立了针对短时间尺度用电行为相关性的网络化分析方法。首先,由描述短时间尺度用电功率序列间相关性的皮尔逊相关系数构造用电行为相关矩阵,为减小随机用电行为与测量误差等无序噪声信息对相关矩阵的影响,引入随机矩阵理论对相关矩阵进行去噪,并提出利用特征值谱熵的正则化修正方法。然后,基于相关矩阵构建了相关矩阵分阶与同级网络序列,提出了基于社团结构的相关性聚类方法与基于最小树的相关性等级结构挖掘方法;定义了量度用电相关性的拓扑指标,从而构造了用电行为相关性聚类分析与等级分析的网络模型,并就模型的典型应用场景进行了讨论。最后,通过实例数据验证了上述去噪方法、相关性分析网络模型及指标的有效性。

Network Model for Correlation Analysis of Short-term Electricity Consumption Behavior

The short-term correlation of electricity consumption behavior is the important decision basis for user meticulous management and intelligent distribution. In this paper, a network method is first proposed to analyze the correlation matrix of electricity sequences. In order to eliminate the effects of noise information from the random behavior and measurement error, random matrix theory is adopted for correlation matrix denoising, and regularization correction exploiting eigenvalue spectrum entropy is also introduced. Then, according to the tiered and peer network sequences constructed from the filtered correlation matrix, a correlation clustering method based on community structure and a hierarchy mining method based on minimum spanning tree are proposed for detailed analysis, including topological indicators to characterize the dynamic evolution of correlation network, as well as the discussion of the model application. Case studies with instance data verify the validity of the proposed denoising method, network analysis model and indicators.