采用动态模式分解分形特征的局部放电模式识别

Partial Discharge Pattern Recognition Based on Fractal Features Using Dynamic Mode Decomposition

为了解决传统局部放电模式识别方法计算量大、识别速度低的问题,本文采用构造二维谱图的方法进行局部放电模式识别。首先,利用动态模式分解算法构造局部放电的二维谱图,以少于构造传统三维谱图的计算量获取局部放电缺陷信号的二维表征。然后,针对不同缺陷信号的二维谱图提取两种分形特征(分形维数和间隙度),且构造了不同缺陷信号下二维谱图的分形特征数据集。最后,对该数据集进行X均值聚类。结果表明,X均值聚类结果优于传统K均值聚类和模糊C均值聚类算法,并且相比于反向传播神经网络和支持向量机算法,本文所提模式识别方法对3种局部放电缺陷信号综合识别率高,算法运算时间短。

To solve the problem of large computation load and low recognition speed with the traditional partial discharge(PD)pattern recognition methods,a method of constructing two-dimensional spectrum is used for the PD pattern recognition.First,the dynamic mode decomposition(DMD)algorithm is used to construct the two-dimensional spectrum of PD for obtaining the two-dimensional characterization of PD defect signals at the cost of a computation load less than that when constructing the traditional three-dimensional spectrum.Then,two kinds of fractal features are extracted from the two-dimensional spectra of different defect signals,i.e.,fractal dimension and lacunarity.In addition,the fractal feature data sets of two-dimensional spectra under different defect signals are constructed.Finally,these data sets are processed by X-means clustering.The result obtained using X-means clustering is better than those using the traditional K-means clustering and fuzzy C-means clustering methods.Moreover,compared with the back propagation neural network and the support vector machine algorithm,the pattern recognition method proposed in this paper has a higher comprehensive recognition rate for three kinds of PD defect signals,as well as a shorter operation time.