基于改进DS证据理论的变压器故障诊断方法

An Approach to Transformer Fault Diagnosis Based on Improved DS Theory

传统证据理论在变压器故障诊断中存在主观局限性,且对证据体可靠性的选取缺乏科学性。为了融合变压器色谱分析数据与电气试验数据,并能全面的反映变压器的状态,文中提出一种基于改进证据理论的变压器故障诊断模型。首先,通过熵权法求出子证据体的相对权重,再结合BP和量子神经网络的优化诊断结果,修正熵权作为证据体的可靠因子。其次,构造子证据体的基本概率分配函数,采用Dempster合成规则实现故障信息融合。最后,将所提诊断方法应用于实际工程案例,诊断结果表明,该诊断方法有效、可行,且提高了诊断准确率。

When traditional evidence theory is used to analyze the fault of transformer,the selection of evidence body lacks scientificity and exists subjectivity.In this paper,an approach based on entropy weight theory is proposed to ensure the reliability of each evidence body,through which the data of Dissolved Gas Analysis and electrical tests data are combined effectively and the operation state of transformer is reflected accurately.Firstly,the relative weight of each evidence body is obtained by entropy method,and then the relative weight is amended by the diagnostic result of BP and quantum neural network.Secondly,evidence is preprocessed through introducing credibility and the basic probability assignment is constructed based on entropy weight.Finally,preprocessed evidence is integrated effectively using Dempster's rule.Through the proposed approach,the multi-characteristic signal is utilized adequately and the diagnostic accuracy is improved,and the practical engineering problems are disposed effectively together with enhancing the ability of distinguishing the uncertainty data.