融合频谱特性与聚类云-证据推理的油纸绝缘老化程度诊断

Diagnosis of oil-paper insulation aging degree by fusing spectral properties and clustered cloud-evidence reasoning

油纸绝缘系统老化程度与电力变压器的稳定运行密切相关。针对油纸绝缘频域介电谱(frequency-domain spectroscopy,FDS)所提相关特征量单一和未考虑频谱特性指标间冲突性与随机性而导致老化诊断结果误差大的问题,提出了融合频谱特性与聚类云-证据推理的油纸绝缘老化程度诊断方法。首先,根据不同老化试样的频谱特征提取老化诊断多特征指标,并利用非线性拟合构建老化特征数据库。其次,基于Kmeans聚类提出适配非线性变化指标的聚类云模型基本概率分配方法。最后,针对多特征量的冲突性及相关性差异,利用CRITIC-G1综合赋权法计算各指标对应的证据修正因子,并进行基本概率再分配,利用证据融合推理得出油纸绝缘系统真实老化程度。实验结果表明,所提方法可准确诊断复合油纸绝缘样本的老化状态,为电力检修策略的制定提供了理论指导。

The aging degree of an oil-paper insulation system is closely related to the stable operation of power transformer equipment.There is low accuracy in aging diagnosis when it relies on a single relevant feature quantity obtained by frequency-domain spectroscopy(FDS)and does not consider the conflict and randomness among the spectral indicators.Thus a new diagnostic method for determining the aging degree of oil-paper insulation is proposed.This method integrates spectral characteristics and clustering cloud-evidence reasoning.First,the multi-feature indicators for aging diagnosis are extracted through aging samples’spectral characterization.An aging feature database is then constructed through nonlinear fitting.Secondly,a basic probability assignment method for the clustering cloud model is proposed based on K-means clustering to accommodate the nonlinear change indicators.Finally,to address the conflict and correlation differences of multi-feature quantities,the CRITIC-G1 comprehensive assignment method is used to calculate the evidence correction factor for each indicator.Subsequently,the basic probability is reassigned,and evidence fusion inference is employed to determine the actual aging degree of the system.The results demonstrate that the proposed method can accurately diagnose the aging state of composite oil-paper insulation samples and offers theoretical guidance for developing power maintenance strategies.