行波暂态高速采集通常采用的低门槛突变量启动,它在确保对弱故障的可靠启动的同时,亦使大量非故障干扰杂波被记录,造成录波数据样本严重不平衡,给多通道故障数据有效筛选带来困难。以母线上多回线路广义电流模量的分形维数及暂态能量作为特征向量,借助历史故障记录及故障后续事件的连续性、时间上的紧密性等知识,以少量已知类别的数据样本作为“锚点”,附加相邻样本的时间邻域约束,构建带条件约束的半监督聚类,形成基于领域知识的数据驱动式行波录波数据分类方法,实现对多通道海量录波数据中故障数据集的有效筛选。大量现场实测数据测试表明,该方法对多通道故障录波数据筛选可行、有效。
Transient high-speed data acquisition of traveling wave is usually set low triggering thresholds. This can ensure it be started in the case of high impedance fault, while some disturbance signals not caused by the fault are also recorded and this make the recorded data seriously imbalanced. It is difficult to effectively screen the contingency data from the imbalanced multi-channel data. Fractal dimension and transient energy of general modal current component for multi- transmission lines on the same bus was selected as characteristic vectors in this paper. With the help of some knowledge, such as the continuity of the historical events records and the following events, the compact support in time domain and so on, the constrained semi-supervised clustering was formed through the "anchor" composed by a small amount of labeled sample set and neighborhood constraints of unlabeled sample set in time domain. And data-driven classification method for traveling wave field data based on domain knowledge was proposed, which makes the screening feasible and effective for massive multi-channel waveform records. The proposed method proved feasible and effective.