146 / 2022-03-14 10:32:03

Power Transformer Defect Prediction Method Based on SMOTE and Random Forest Algorithm

Power transformer,Defect prediction,Smote,Unbalanced dataset,Random Forest

Purpose/Aim

The defect data in the transformer data is far less than the non-defect data, which leads the supervised learning model to pay more attention to the non-defect data, which makes the model perform poorly on the defect data.  

Experimental/Modeling methods

This paper firstly uses the synthetic minority oversampling technique (SMOTE) algorithm to supplement the number of defect data samples. Then the transformer defect prediction process is regarded as a classification process, and in order to find the most suitable machine learning model for power transformer defect prediction, we used machine learning models such as decision tree, logistic regression, SVM, random forest, etc.

Results/discussion

After oversampling the original sample set with the SMOTE algorithm, the Kappa coefficients of all classification algorithm models are improved; the Kappa coefficients of the random forest model are ahead of other algorithms mentioned in this paper.

Conclusions

A power transformer defect prediction method based on SMOTE algorithm and random forest algorithm is proposed. The SMOTE algorithm is used to alleviate the class imbalance of transformer data, which can effectively improve the classification performance of the model. The classification performance of the random forest algorithm is better than other classification algorithms under the condition of the sample set in this paper, and it realizes the acquisition of transformer defect prediction knowledge, which can help the inspection personnel to extract key features and evaluation rules from the acquired data.