CAJ | 학술논문

为实现GIS故障检测和缺陷的模式识别，针对GIS出现的典型绝缘缺陷及其放电特点，设计了4种常见的GIS绝缘缺陷模型并进行放电试验。对获取的大量超高频包络信号，提出了一种基于小波包变换奇异值分解的时域特征提取方法。该方法首先对包络信号进行小波包变换，构建各尺度的小波包分解系数矩阵，然后对其进行奇异值分解，提取特征向量，在此基础上，采用BP神经网络进行模式识别，结果表明采用此方法获得了良好的识别效果。
위실현GIS고장검측화결함적모식식별，침대GIS출현적전형절연결함급기방전특점，설계료4충상견적GIS절연결함모형병진행방전시험。대획취적대량초고빈포락신호，제출료일충기우소파포변환기이치분해적시역특정제취방법。해방법수선대포락신호진행소파포변환，구건각척도적소파포분해계수구진，연후대기진행기이치분해，제취특정향량，재차기출상，채용BP신경망락진행모식식별，결과표명채용차방법획득료량호적식별효과。
In order to achieve the GIS fault detection and defect type recognition, four typical defect models were designed and discharge tests are carried out aiming at insulation defect as well as discharge characteristics in the GIS . With a large number of ultra high frequency envelope signal, a method of do-main feature extraction was proposed based on wavelet packet transform with singular value decomposi-tion. The envelope signal was decomposed through wavelet packet transform first in the method, then the coefficient matrix of wavelet packet transform was built in the scale, after that feature vectors of matrix were extracted by means of singular value decomposition. On this basis, BP neural network was took ad-vantage of for pattern recognition . The results show that the good recognition effect is obtained with that method .