基于波形突变特征的海量行波数据中雷击干扰波形辨识  

Identification of Lightning Interference Waveforms in Massive Traveling Wave Data Based on Waveform Mutation Characteristics

作　　者：曹璞璘[1] 申忠友 束洪春[1] 马御棠 韩一鸣 朱梦梦 CAO Pulin;SHEN Zhongyou;SHU Hongchun;MA Yutang;HAN Yiming;ZHU Mengmeng(Faculty of Electric Power Engineering,Kunming University of Science and Technology,Kunming 650500,China;Luzhou Power Supply Company,State Grid Sichuan Electric Power Company,Luzhou 646000,China;Electric Power Research Institute of Yunnan Power Grid Co.,Ltd.,Kunming 650217,China)

机构地区：[1]昆明理工大学电力工程学院,云南省昆明市650500 [2]国网四川省电力公司泸州供电局,四川省泸州市646000 [3]云南电网有限责任公司电力科学研究院,云南省昆明市650217

出　　处：《电力系统自动化》2023年第17期178-186,共9页Automation of Electric Power Systems

基　　金：国家自然科学基金资助项目(52167012,51977102);云南省基础研究计划资助项目(202001AT070096)。

摘　　要：行波装置等高采样率装置已在电力系统得到普遍应用,未来将成为新型电力系统运行数据积累的重要来源。行波装置不仅会记录故障产生的电磁暂态过程,也会记录雷击干扰等其他扰动。雷击干扰虽然会引起绝缘的部分损伤,但由于缺少跳闸记录等事件记录的支撑,且高采样率数据存在离群点问题,雷击识别存在较大困难。文中提出基于中位数绝对离差法滑动窗口的行波数据预处理方法,消除绝大部分异常坏点并对其暂态细节进行修复,抑制异常坏点引起的波形奇异点,对预处理后行波数据做形态学梯度提取,抑制波形中变化较为平缓部分,放大局部暂态突变细节,从而提取电流行波数据突变的相关特征。采用合成少数类过采样技术平衡少数类样本,利用随机森林无需特征归一化的特点,避免新加入数据造成归一化重置的问题,实现了海量数据中线路扰动数据的初步提取,并通过雷电定位系统完成验证。High sampling-rate devices such as traveling wave devices have been widely used in power systems and will become an important source of accumulated operational data for new power systems in the future.Traveling wave devices not only record the electromagnetic transient process generated by faults,but also record other interference such as lightning interference.Although lightning interference can cause partial damage to the insulation,it is difficult to identify lightning strikes due to the lack of support for event records such as trip records and the presence of outliers in high sampling-rate data.The paper proposes a preprocessing method for traveling wave data based on the median absolute dispersion sliding window method,which eliminates the majority of abnormal bad points and repairs their transient details,suppresses waveform singularities caused by abnormal bad points,performs morphological gradient extraction on the preprocessed traveling wave data,suppresses the relatively gentle changes in the waveform,amplifies local transient mutation details,and extracts relevant features of current traveling wave data mutations.The synthetic minority oversampling technique is used to balance minority-class samples,and the problem of normalization caused by newly added data is avoided by using the characteristics of random forest without feature normalization.The initial extraction of line disturbance data from massive data is realized,and the verification is completed by lightning positioning system.

关 键 词：电磁暂态 雷击干扰 行波装置 形态学梯度 随机森林 

分 类 号：TM863[电气工程—高电压与绝缘技术]