基于电压一阶差分离群点关联分析的配电网线变关系诊断  

Diagnosis of feeder-transformer connectivity in a distribution network based on voltage first-order difference outlier correlation analysis

作　　者：陈锦铭 蒋玮[1] 袁宇波 曾飞 卢庆宁 徐俊俊 CHEN Jinming;JIANG Wei;YUAN Yubo;ZENG Fei;LU Qingning;XU Junjun(School of Electrical Engineering,Southeast University,Nanjing 210096,China;Electric Power Research Institute,State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 211103,China;College of Automation&College of Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]东南大学电气工程学院,江苏南京210096 [2]国网江苏省电力有限公司电力科学研究院,江苏南京211103 [3]南京邮电大学自动化学院/人工智能学院,江苏南京210023

出　　处：《电力系统保护与控制》2024年第18期158-168,共11页Power System Protection and Control

基　　金：国家重点研发计划项目资助(2022YFB2404200);国家自然科学基金项目资助(52277172)。

摘　　要：基于馈线-配变电压相关性计算及其扩展方法,是实现配电网线变关系诊断的有效手段。然而低压配网运行工况复杂,配变电压在部分时段异常波动,导致该类方法存在一定局限性。为此,在电压相关性方法上进行拓展,利用电压突变特征随潮流扩散机理,提出了基于电压一阶差分离群点关联分析的线变关系诊断方法。首先,基于电压相关性方法对配变进初筛,得到可信与异常配变集。其次,运用单变量时间序列流式方法开展电压一阶差分值离群点检测,并与变电站调压事件关联实现母线离群点标签化。随后,从可解释性、可感知性和可区分性3个维度对离群点进行评价,形成母线带权离群点序列。最后,提出了基于离群点匹配度的线变关系诊断方法,通过横纵向综合比较推荐异常配变所属馈线。实验结果表明,所提的线变关系诊断方法优于传统方法,可有效克服复杂工况的影响,显著提升诊断算法的性能。The voltage correlation calculation and its extension method is an effective means to realize the diagnosis of feeder-transformer relationship in distribution networks.However,due to the complex operating conditions of low-voltage distribution networks,the distribution transformer voltage fluctuates abnormally in some periods,which leads to certain limitations of this type of method.For this reason,this paper expands on the voltage correlation method,utilizes the mechanism of voltage mutation feature diffusion with the trend,and proposes a feeder-transformer relationship diagnosis method based on the first-order difference separation group point correlation analysis of voltage.First,based on the voltage correlation method,the initial screening of the distribution transformers is carried out to derive the credible and abnormal distribution transformer sets.Secondly,a univariate time series streaming method is applied to carry out voltage first-order difference outlier detection,and the bus outlier labeling is realized by correlating with the substation voltage regulation events.Subsequently,the outlier points are evaluated from three dimensions of interpretability,perceptibility and distinguishability to form a sequence of bus weighted outlier points.Finally,a feeder-transformer relationship diagnosis method based on the matching degree of outlier points is proposed to recommend the feeder to which the anomalous distribution transformer belongs through a comprehensive horizontal and vertical comparison.Experimental results show that the feeder-transformer relationship diagnosis method proposed in this paper can effectively overcome the influence of complex working conditions and significantly improve the performance of the diagnosis algorithm.

关 键 词：配电网 线变关系诊断 电压波动 离群点检测 

分 类 号：TM73[电气工程—电力系统及自动化]