矢量量化在局部放电模式识别中的应用

Application of Vector Quantization to Partial Discharge Pattern Recognition

利用绝缘试品在升、降压过程中放电量随电压变化构成的视在放电量–施加电压模式序列作为局部放电特征量,并将矢量量化和快速匹配算法引入局部放电模式识别的研究中。该算法在分类器训练阶段首先利用训练样本集通过LBG编码技术构造码书,再分别对各类放电的训练样本进行矢量量化编码,计算码字频率矩阵;在测试阶段,以同样的流程对待识别样本进行矢量量化编码和码字频率矩阵计算。最后将训练样本和待识别样本的码字频率矩阵利用快速匹配算法进行匹配后得到识别结果。对5类放电的100个样本的检测结果表明,该算法具有执行简便、识别率高的优点。

The data sequences of apparent charge versus applied voltage （AQ-U） pattern is used as characteristic features of partial discharge. A novel method based on vector quantization （VQ） technology is introduced to realize partial discharge （PD） pattern recognition. In training process of classifier, this method firstly makes use of LBG encoding technique to design codebook. The training samples of different PD sources are then encoded according to the codebook. After that, the code occurrence frequency for each PD source is calculated. In testing process, the same procedure is applied to test samples for vector quantization and code occurrence frequency calculation. Finally, the recognition results are obtained by fast matching the code occurrence frequency matrixes of training and testing samples. The recognition results of 100 samples of five PD sources demonstrate that such classifier has the advantages of simple implementation and high classification rates.