基于多信息融合的变压器故障诊断

Fault Diagnosis of Transformer Based on Multi-information Fusion

目前变压器故障诊断中,以油色谱数据为主的诊断方法信息量不足,同时,基于单一智能算法的故障判断能力有限。鉴于此,文中构建了一种基于多信息融合的变压器故障诊断模型。该方法首先将电气试验等数据与17组油中特征气体含量比值相结合作为故障特征量,以获得更丰富的故障信息,并采用特征敏感性评估与核主元分析方法对所选定的故障信息进行特征降维融合,以实现多层面故障信息的互补。其次,将降维融合后的9维特征量分别作为BP神经网络、SVM及贝叶斯方法的特征输入,进行故障类型的初步判定。最后,若初步诊断结果不存在分歧,则直接得出结论;否则,基于证据理论方法分别计算各证据体的基本信任分配,并运用Dempster组合规则进行决策融合,以做出更合理的故障判断。实例分析表明,该方法有效解决了信息单一和方法单一的问题,且有效提高了故障识别准确率。

For current fault diagnosis of power transformer, the method based on dissolved gas chromatographic data analysis is insufficient in fault information, and its diagnostic capability based on single algorithm is limited. This paper proposes a transformer fault diagnosis method based on multiple information fusion. The method firstly combines the data of electrical test with 17 groups of the concentrations of characteristic gas components in oil to achieve more fault information. The feature sensitivity evaluation and the kernel principal component analysis（KPCA） are used to reduce the feature dimension of the selected fault data to realize complementary of the multi-level fault information. Then, the nine-dimensional features after dimension reduction are taken as the input characteristic parameters of the BP neural network, the support vector machine （SVM） and the Bayesian method to determine fault type. Subsequent- a conclusion is output if there is no disagreement on the preliminary diagnosis result, otherwise, the basic belief assignment of each evidence is recalculated and the Dempster combination rule is used for decision fusion to make more reasonable fault judgment. Example analysis shows that the proposed method can effectively deal with the problems of insufficient information and single method, and improve diagnostic accuracy.