基于CLAHE增强的GIS局放脉冲序列识别算法

Recognition Algorithm of GIS Partial Discharge Phase Resolved Pulse Sequence Based on CLAHE Enhancement

为了研究气体绝缘组合电器(gas insulated switchgear,GIS)的特高频局部放电脉冲序列识别算法,进而提高其绝缘故障诊断的正确率,提出利用限制对比度自适应直方图均衡化(contrast limited adaptive histogram equalization,CLAHE)算法对脉冲序列分布(phase resolved pulse sequence,PRPS)图谱进行预处理,实现放电脉冲目标的强化,从而增强数据集视觉特征的方法;然后计算增强图谱的均匀局部二值模式(uniform local binary pattern,ULBP)作为特征向量,利用Adaboost级联分类器不断提高识别率直至收敛,从而实现GIS内部绝缘故障类型的识别。实验结果表明:CLAHE增强将识别率的上限从93.36%提高到了96.09%;在变化的外施电压下,ULBP特征向量比传统图像特征的识别率提高了10.71%~15.72%;Adaboost强分类器在训练时对样本数量的要求降低了约1/3。故所提算法进一步扩大了优化空间,增强了传统算法的泛化能力,提高了训练效率。

A novel UHF partial discharge phase resolved pulse sequence(PRPS)recognition algorithm for gas insulated switchgear(GIS)is proposed in this paper in order to improve the accuracy of insulation fault diagnosis.Firstly,the PRPS pattern is preprocessed by contrast limited adaptive histogram equalization(CLAHE)algorithm,which can enhance the discharge pulse target and thus enhance the visual features of the data set.Afterward,the uniform local binary pattern(ULBP)is calculated as the eigenvector of pattern to be recognized.Finally,an Adaboost cascade classifier is designed for gradually increasing the recognition rate up to convergence.Experimental results show that CLAHE enhancement increases the upper limit of recognition rate from 93.36%to 96.09%;the recognition rate of ULBP eigenvectors is 10.71%-15.72%higher than that of traditional image features under variable applied voltage.The Adaboost strong classifier reduces the requirements of sample size by about 1/3 during training.As a result,the algorithm proposed further expands the optimization space,enhances the generalization ability of the traditional algorithm and improves the training efficiency.