【摘要】 人工智能的快速发展为变压器的故障诊断提供了准确率更高的新方法，但是现有的故障诊断模型不利于处理不平衡数据集。为提高变压器故障诊断的准确率，提出利用合成少数类过采样技术（synthetic minority oversampling technique,SMOTE）和随机森林相结合的诊断方法，利用SMOTE算法对变压器油色谱故障数据集的少数类故障样本进行扩充，以平衡各个故障类型数据的数量。随后使用随机森林分类器分别对未经扩充和经SMOTE扩充的数据进行故障识别，研究两者结合使用的效果。诊断结果表明，使用SMOTE对不平衡变压器油色谱故障数据集进行扩充后再进行故障诊断，可以显著提高故障诊断的准确率。另外还分析其他两种故障诊断模型的结果，验证上述结论的同时，得出随机森林分类器是3种故障诊断模型中诊断准确率最高的模型，为变压器进行故障诊断提供一种较为理想的方法。更多还原

【Abstract】 The rapid development of artificial intelligence provides a new method with higher accuracy for transformer fault diagnosis,but the existing fault diagnosis models are unfit to handle imbalanced datasets.In order to improve the accuracy of transformer fault diagnosis,a diagnosis method combining synthetic minority oversampling technique(SMOTE)and random forest was proposed.The SMOTE algorithm aimed to expand the minority fault samples of transformer oil chromatography fault datasets to re-balance the number of data of each fault type.We analyzed the effect on combining SMOTE and random forest.Then,the random forest classifier was used to identify the faults of the data that had not been expanded and the data that had been expanded by SMOTE respectively.The diagnosis results indicated that the accuracy of fault diagnosis can be significantly improved by using SMOTE to expand the unbalanced transformer oil chromatography fault data set before fault diagnosis.In addition,several other fault diagnosis models,were investigated in order to illustrate the effectiveness compared to their results of the proposed model.It is concluded that the random forest classifier is proved with the highest diagnostic accuracy among three fault diagnosis models,providing an ideal method for transformer fault diagnosis.更多还原