基于离散隐式马尔科夫模型的局部放电模式识别

摘要

In this paper,the data sequences of apparent charge versus applied voltage（ΔQ-U） in the process of stepping-up/down voltage is used as characteristic features for pattern recognition of partial discharge（PD）.Discrete hidden Markov models（DHMMs） classifier is introduced to realize the PD pattern recognition.Firstly,by utilizing vector quantization method,a codebook is formed based on LBG encoding data,and then the codebook index sequences are assigned to the train and test samples of various PD types respectively.In the training of the classifier,the DHMMs are obtained for each PD source.In the testing process,the output probabilities of the test samples in all DHMMs are calculated.The model number with the largest probability is chosen as the classification results.The recognition results from 5 PD sources and 150 samples demonstrate high classification rates and easy expansion of the classifier.%利用绝缘试品在升、降压过程中的视在放电量-施加电压序列作为局部放电特征量,并将离散隐式马尔科夫模型分类器引入局部放电模式识别的研究中。该算法首先利用矢量量化方法通过对码本生成样本集进行LBG编码构造码本,并分别对各类放电的训练和测试样本分配码本索引序列。在分类器的训练阶段,输入训练样本序列训练得到每类放电的离散隐式马尔科夫模型。在测试阶段,计算每类离散隐式马尔科夫模型输出测试样本序列的概率,取最大概率对应的模型序号作为识别结果。对5类放电的150个样本的识别结果表明,离散隐式马尔科夫模型具有识别率高、易扩展的优点。