摘要
针对变压器故障样本类别不平衡和模型诊断精度低的问题,首先使用自适应综合过采样对训练样本中少数类别进行扩充以平衡样本,然后通过深度信念网络对平衡后样本进行深层特征提取,最后将特征向量输入到XGBoost(extreme gradient boosting)进行故障诊断。算例分析表明,所提出的诊断模型准确率最高达91.94%;在样本类别不平衡条件下,所提故障诊断方法与BP神经网络、支持向量机、随机森林、XGBoost相比更优。
Aiming at the problem of unbalanced categories of transformer fault samples and low diagnostic accuracy of the model,adaptive comprehensive oversampling is used to expand the minority categories in the training samples to balance the samples,and then the deep belief network is used to extract the deep features of the balanced samples.The eigenvectors are input to XGBoost(extreme gradient boosting)for fault diagnosis.The analysis of the example shows that the accuracy of the proposed diagnosis model is up to 91.94%.Under the condition of unbalanced sample categories,the proposed fault diagnosis method is better than BP neural network,support vector machine,random forest and XGBoost.
作者
刘可真
梁玉平
王科
赵勇军
LIU Kezhen;LIANG Yuping;WANG Ke;ZHAO Yongjun(Faculty of Electric Power Engineering,Kunming University of Science and Technology,Kunming 650500,China;Electric Power Research Institute,Yunnan Power Grid Co.,Ltd.,Kunming 650217,China;Yunnan Electric Power Technology Co.,Ltd.,Kunming 650000,China)
出处
《电力科学与工程》
2022年第11期9-16,共8页
Electric Power Science and Engineering
基金
国家自然科学基金(51477100)
云南电网有限责任公司科技项目(YNKJXM20180736)。
