Fault Diagnosis for Oil-immersed Transformer Based on Missing Data Imputation
Data quality is an important factor affecting the accuracy and reliability of transformer fault diagnosis models.Aiming at the existing transformer fault diagnosis model with higher requirements for data integrity,we proposed a fault diagnosis method based on missing data imputation for oil-immersed transformers.Firstly,the missing data of transformer samples were filled by using the extremely randomized trees(ERT),and the predictive effect of ERT model was evaluated by comparing with various regression models.Then,a 16-dimensional feature set representing operating status of trans-formers was extracted based on the dissolved gas data in oil,and the transformer fault diagnosis samples with complete information were obtained.Finally,the tree-structure probability density estimation(TPE)algorithm was used to achieve the parameter optimization of the gradient boosting decision tree(GBDT)model,and a transformer fault diagnosis model based on TPE-GBDT was constructed.The results show that,when filling the transformer sample data with a missing rate of 10%,the coefficient of determination of the ERT algorithm reaches 0.96,which is higher than that of the algorithms such as linear regression and random forest regression.Moreover,the average diagnostic accuracy and standard deviation of the TPE-GBDT model based on the ERT imputed sample data are 90.1%and 0.036,respectively,which are superior to those of the algorithms such as linear discriminant analysis and random forest classification.This method can be adopted to effectively improve the transformer sample quality and the fault diagnosis effect,which can provide targeted guidance suggestions for transformer operation and maintenance.
transformermissing data imputationextremely randomized treesfault diagnosisgradient boosting treedissolved gas analysis
万方数据
维普
