摘要
在气体绝缘组合电器(GIS)实体模型内部分别放置针-板、悬浮金属颗粒和绝缘子表面固定金属颗粒三种缺陷模型,用超声波传感器在相同电压下采集到良好的局部放电波形,将从现场运行设备上测得的背景噪声叠加到原放电波形上。对叠加噪声后的放电波形采用小波去噪,针对波形特点选取了7个特征参数,分别用去噪前后波形的特征参数对BP_Adaboost分类器进行训练和测试,结果表明用去噪后波形提取的特征量作为分类器输入的识别率更高。
Needle-plate, suspended metal insulator surface were placed separately in GIS particles and metal particles fixed on entity model. The discharge waveforms were detected by using ultrasonic sensor under the same voltage. In order to be consistent with the field noise, we added the noise which was detected from the field equipment to the discharge waveforms. The waveforms were processed by wavelet de-noising. Then Aiming at the waveforms＇ chara- cteristics, seven characteristic parameters were chosen. The characteristic parameters before and after de-noising were used separately to train and test BP Adaboost classifier. The results showed that by using characteristic vectors grasped from waveforms after de-noising, the recognition result is higher.
出处
《电气工程学报》
2017年第11期41-45,共5页
Journal of Electrical Engineering
