基于原子稀疏分解的配电网故障选线机理研究

Bearing in mind the inaccuracy and low reliability of existing fault line selection methods owing to their lack of feature information caused by their not simple and sparse enough characteristic components, a new selection method based on atomic sparse decomposition is introduced in this study which includes the following aspects: ① Considering the zero sequence current signal characteristics of fault line and the differences and characteristics of time-frequency distribution of existing atoms, a zero sequence current atomic model is established; the corresponding relationship between the atom and its characteristic component is then set up through model identification to collect the feature model atoms; finally the unified selection of self-adjusted solution of different fault feature bands is achieved. ② With the research on replacement method of traditional “the biggest inner product” matching selection principle, correlation algorithm is brought into waveform matching degree measurement; based on genetic and particle swarm algorithm, the traditional matching pursuit search is optimized. ③ With the acquisition of transient characteristic atomic components, research on single symptom domain diagnosis and information fusion fault detection of DS evidence theory is made as well as fault line selection voting mechanism of atomic energy spectrum entropy algorithm, the preliminary vote of fuzzy clustering and fusion vote. Besides the technical support for the development of technology of intelligent distribution network, the research will also provide theoretical support for the accurate and reliable real-time fault line selection.

针对基分解方法所得特征分量不够简洁、稀疏导致的特征信息欠缺，进而造成故障选线准确率和可靠性不高等问题，提出一种基于原子稀疏分解机理的配电网故障选线方法，主要内容包括：①针对故障零序电流信号特点，结合现有原子的时频分布差异及特性，建立零序电流原子库模型；采用模态辨识方法，建立原子与特征分量之间的对应关系，提取出特征模态原子；研究不同类型线路故障特征频带统一选取的自适应解决方案。②开展传统“内积最大”匹配选取原则的替代方法研究，将互相关运算等方法引入到衡量波形匹配度上；在遗传、粒子群等算法基础上，开展传统匹配追踪搜索方式的优化工作。③在获取暂态特征原子等分量基础上，开展单征兆域诊断、DS证据理论的信息融合故障判据研究；进行原子能谱熵计算、模糊聚类的初步投票以及融合投票的故障投票选线机制研究。本课题的研究成果将为准确、可靠的实时故障选线提供重要的理论支撑，并为智能配电网技术的发展提供技术支持。

本课题针对基分解方法所得特征分量不够简洁、稀疏导致的特征信息欠缺，进而造成配电网故障选线准确率和可靠性不高等问题，提出了一种基于原子稀疏分解机理的配电网故障选线方法，主要研究内容包括：①针对故障零序电流信号特点，结合现有原子的时频分布差异及特性，建立了零序电流原子库模型；采用模态辨识方法，建立了原子与特征分量之间的对应关系，提取出了特征模态原子；研究了不同类型线路故障特征频带统一选取的自适应解决方案。②开展了传统“内积最大”匹配选取原则的替代方法研究，将互相关运算等方法引入到衡量波形匹配度上；在遗传、粒子群等算法基础上，开展了传统匹配追踪搜索方式的优化研究工作。③在获取暂态特征原子等分量基础上，开展了单征兆域诊断、DS证据理论的信息融合故障判据研究；进行了原子能谱熵计算、模糊聚类的初步投票以及融合投票的故障投票选线机制研究。④针对高阻接地故障时，故障电流与正常负荷电流相差无几，故障零序电流微弱等导致的现有选线方法难以适应等问题，初步开展了随机共振理论在故障选线领域的研究工作。具体取得的重要结果包括：①建立了衰减正弦量原子模型，改进了灰色关联分析理论，准确测度了故障特征。②遴选出了故障特征最佳原子，搭建了极限学习机故障诊断模型。③采用DS证据融合了各故障原子特征，建立了多特征原子综合信任度判据。④构建了特征原子奇异熵的综合测度故障选线判据。⑤搭建了适合故障暂态零序电流精确特征提取的混合原子库。⑥提出了一种解决强噪声背景下微弱暂态零序电流检测新方法。在本课题的研究过程中，共培养硕士4名，本科生2名；在国际、国内学术期刊及会议上累计发表论文34篇，其中，SCI收录8篇、EI期刊收录19篇；授权发明专利11项、申请发明专利1项；获省部级奖励3项。本课题的研究成果可解决固定基分解存在的特征分量不简洁、不稀疏等问题，可为准确、可靠的故障选线提供理论支撑，并为智能配电网技术的发展提供技术支持。