基于选择性聚类融合的小电流接地系统故障选线方法

Many power distribution systems have been operated with neutral unearthed, high resistance earthed or resonance earthed. They are usually called neutral ineffectively earthed power systems. When the single-phase faults were present, over-voltage would develop. If it is not eliminated immediately, it will bring about cable bombing and other faults. The safety of the power system and the quality of the electricity will be affected. Therefore, earth fault feeder detection is a long-standing problem in neutral ineffectively earthed power systems. Traditional ground fault detection schemes has poor precision and reliability because of minor fault current, fault arc, harmonics influence. For improvement, a novel principle for fault feeder detection based on clustering ensemble selection algorithms is presented in the proposed project. First, Copula and multivariate GARCH model are applied to analyze correlation between multi-sources fault information, which can exactly extract the fault features. Then, time-space distribution characteristics are analyzed to identify data configuration. What's more, clustering ensemble selection algorithms are employed to fuse all kinds of fault information and integrate different single clustering algorithms. Finally, two kinds of fault feeder detection schemes are proposed. One is for all feeders, which can classify all feeders into two groups depending on their fault features by clustering ensemble selection algorithms. The other is just for one feeder, and different period of time before and after grounding fault occurring are divided into two groups depending on their fault features by clustering ensemble selection algorithms. The space relative distance among detected patterns and two cluster centers is then calculated to discriminate the faulted feeder. Simulation and experimentation will be applied to test the prototype of scheme. The proposed project can avoid the influence due to the variations of system operation mode, and inherent limitation of the single protection scheme can be eliminated and the precision and robustness of fault detection can be improved.

小电流接地系统故障选线是我国在建、改建的配电系统亟待解决的重要课题。但由于故障残流小、故障电弧及干扰影响等原因，现有方法在实际运行中选线正确率很低。本项目深度挖掘故障信息以提高信噪比，并充分利用选线技术之间、多种融合算法之间的互补性，提出基于选择性聚类融合的选线方法，以提高选线正确率。首先借助Copula函数和多元GARCH模型等统计方法，刻画多源故障信息之间的相关性，提取准确、充分的故障信息；借鉴统计物理学思想，将故障特征量投影到多维空间，引入具有明确物理机理的选择性聚类融合算法区分故障信息，利用空间相对距离构造选线判据；通过配电网多个保护单元彼此之间检测到的接地故障信息特性的不同，构建基于多保护单元的选线方法；通过单一保护单元内部不同选线方法在不同历史时段所呈现出的特性不同，构建面向单一保护单元的选线方法。该项目提出的选线方法不受系统运行方式变化的影响，有望提高配电网故障选线的精确度。

小电流接地系统故障选线是长期存在的难题。为深入挖掘故障信息的本质特征和内在规律，对多源故障信息进行有机综合处理，本项目提出了一系列配电网单相接地故障保护方法。首先，为了更有效的提取故障特征量，分析了故障数据的空间分布结构，借助Copula函数刻画了多源故障信息之间的相关性，通过最大似然法从候选的copula函数中选择出与故障特征数据拟合度最好的最优copula函数；而后，由最优copula函数构造秩相关系数，并用秩相关系数取代主成分分析中的线性相关系数，构造出改进的主成分分析法，从而能更好的刻画故障信息特性。其次，引入具有明确物理机理的选择性聚类融合算法区分故障信息，由于聚类的结果会受到样本数量、均匀程度、分布结构等因素影响，因此在实际应用中，针对故障选线的数据结构，揭示和寻求数据的内在结构和规律，进而选用或构造合适的聚类方法，本项目探讨了样本相似性度量和聚类分析的基本原理，将模糊聚类、系统聚类、判别分析等统计物理学思想应用到配电网保护，并给出定量判定聚类有效性的方法，寻找出适用于馈线接地保护的最佳聚类算法。最后，利用空间相对距离构造选线判据；通过配电网多个保护单元彼此之间检测到的接地故障信息特性的不同，构建基于多保护单元的选线方法；通过单一保护单元内部不同选线方法在不同历史时段所呈现出的特性不同，构建面向单一保护单元的选线方法。此外，本项目进一步将聚类算法应用到非有效接地电网的发电机保护当中，取得了较好的效果。该项目提出的保护方法不受系统运行方式变化的影响，有望提高配电网故障选线的精确度。