基于多特征信息融合技术的局部放电模式识别研究

Pattern Recognition for Partial Discharge Based on Multi-feature Fusion Technology

针对单一特征信息分析模式独立辨识不能有效、可靠地判断出局部放电(PD)类型致使识别"误诊"的问题,以及为了最大限度地利用特高频(UHF)传感器所获取的丰富绝缘状态信息,利用放电时间、放电相位分布及UHF能量与放电量相关性等3类特征信息的共性和差异性进行融合互补,提出一种基于多特征信息融合的PD模式识别方法。通过在一套三相分箱式真实GIS(ZF-10-126)试验平台上实测所得的PD试验信息进行分析其结果表明:3类特征信息独立辨识各类缺陷的准确度存在较大差异性和不确定性,但有着各自优势。而采用D-S证据理论进行3类特征信息融合PD模式识别技术,可对3类特征信息独立识别法各自存在的不足进行互补,具有更高更准确的辨识率和可靠性。至此验证了所提方法的有效性与正确性。

The independent identification by analysis pattern of single feature cannot effectively and reliably determine the partial discharge（PD） type,leading to the problem of recognition ＂diagnosis＂. In order to maximize use of rich information of insulation condition which is acquired by the UHF method, combining the differences of the time resolved partial discharge（TRPD）, the phase resolved partial discharge（PRPD） and the relationship between UHF signal energy and charge quantity, we proposed a pattern recognition method which is based on multi-features fusion technology. The PD testing information accessed from the PD experiment which was carried out on a real three-phase 110 k V GIS（ZF-10-126） was adopted to analyze the experimental results. The results show that the difference and uncertainty for four defects which are designed in this paper exist in the recognition of single feature. While after using the similarity and complement of three features, the recognition has a higher recognition rate and certainty by the D-S evidence theory. Moreover, the advantages and effectiveness of the proposed method are verified by the experimental results.