【摘要】 油纸绝缘作为电力变压器的主要绝缘方式,气隙放电特别是大气隙时的放电对其危害极大。为全面评估气隙放电的危害程度,制作了五种气隙尺寸的放电模型,研究气隙大小对局部放电特性的影响,并通过试验研究了其放电发展过程中不同放电阶段的局部放电信号,基于聚类-随机森林算法实现了不同气隙尺寸的气隙放电发展阶段识别。研究结果表明:相对于小气隙缺陷的放电,大气隙放电起始放电场强低,起始放电量大,起始放电相位滞后,当放电发展到后期,大气隙中正放电脉冲的相位不会发展到工频负半周;大气隙缺陷内较小的气隙表面电子脱陷概率和较小的反向电场是使其起始放电相位滞后的主要原因;聚类分析将大小气隙模型的放电分为三个阶段:初始放电阶段、微弱放电阶段和放电爆发阶段;对于发展阶段的识别,相比于径向基函数(RBF)神经网络和核函数支持向量机,随机森林的识别准确率更高,达到93.15%。试验结果为更准确地评估气隙放电的发展阶段提供了依据。更多还原

【Abstract】 Oil-paper insulation is commonly used in power transformer. Air-gap discharge, especially the discharge in large cavity will threaten the oil-paper insulation. In order to comprehensively evaluate the risk of air-gap discharge, five types of cavity configurations were manufactured, and the effects of their cavity sizes on partial discharge(PD) were studied. Then PD signals throughout the accelerated deterioration experiments were analyzed. In addition, based on Clustering-Random Forests, PD development stages of large and small cavities were recognized. Results show that, compared to small cavity PD, large cavity PD possesses lower inception field, higher charge magnitude and higher inception phase, moreover when air-gap PD comes to the last stage, positive PD in large cavity fails to expand to the negative half cycle and vice versa. Lower surface electron emission rate and lower reverse field in the large cavity are the main reasons of higher inception phase. Through clustering, PD development stages for large and small cavity models are both divided into three stages, i.e. initial discharge stage, weak discharge stage and outbreak discharge stage. For the development stage recognition, the accuracy of Random Forests is 93.15%, showing better performance than those of RBF Neural Network and Kernel Based Support Vector Machine. Experiment results provide reference for more precise evaluation of the air-gap PD development stage.更多还原