基于NSST域增强的直流XLPE电缆局放识别方法

Partial Discharge Recognition Method of DC XLPE Cable Based on NSST Domain Enhancement

针对直流交联聚乙烯(XLPE)电缆局放信号图的对比度差和边缘清晰度低,影响电缆缺陷识别率的问题,提出一种基于非下采样剪切波变换(NSST)域增强的直流电缆局放识别方法。首先采用NSST对不同缺陷类型的电缆局放信号图进行分解,得到低频子带图和高频方向子带图,其中,低频子带图进行布谷鸟优化多尺度Retinex算法(CS-MSR)增强处理,高频方向子带图进行亮度保持动态模糊直方图均衡算法(BPDFHE)增强处理。然后对NSST各子带图进行加权重构,得到增强的直流电缆局放信号图。最后提取直流电缆局放信号图中每个子带图的最大放电次数、平均放电时间间隔、平均放电量等共计72个特征参数,并代入到线性核支持向量机(L-SVM),高斯核支持向量机(G-SVM),多项式核支持向量机(P-SVM)和多核支持向量机(M-SVM)识别模型中进行分析。实验结果证明:基于NSST域增强的直流电缆局放信号图与原始信号图相比,在M-SVM情况下,整体缺陷识别准确率提高9.46%以上。运用新提出的方法进行局放识别时,信号图细节更加丰富,且提高了电缆局放信号图的缺陷识别率,为直流电缆局放缺陷识别提供了新的思路。

The contrast ratio and the edge resolution of partial discharge （PD） image are poor in DC XLPE cable, which affect the recognition rate of the defect type. To solve this problem, we proposed a PD recognition method of DC XLPE cable based on non-subsample shearlet transform （NSST） domain enhancement. Firstly, the NSST was used to decompose the PD image of different defect types. The low-frequency subband image and high-frequency direction subband images were obtained. The low-frequency subband image was enhanced by multi-scale Retinex method based on cuckoo search optimization （CS-MSR）, and high-frequency direction subband images were enhanced by brightness preserving dynamic fuzzy histogram equalization method （BPDFHE）. Then, the weighted NSST subband images were reconstructed, and the enhanced PD image was obtained. Finally, the maximum number of discharges, the average discharge time interval and the average discharge quantity, etc., a total of 72 characteristic parameters in the PD images were extracted. The features were brought into linear kernel support vector machine （L-SVM）, Gaussian kernel support vector machine （G-SVM）, the polynomial kernel support vector machine （P-SVM） and multi-kernel support vector machine （M-SVM） for analysis. The experimental results show that the overall recognition accuracy of enhanced DC cable PD image based on the NSST do- main is improved by 9.46% than the original PD image in M-SVM. By using the proposed method, the details of DC cable PD image are more abundant than those of the original PD image, and the defect recognition rate of the PD image is improved, which provides a new idea for PD defect identification in DC cable.