Transformer Defect Prediction Based On Imbalanced Data Set

With the improvement of the power system,the safety and stability of power transmission equipment has attracted more and more attention from scholars and the general public.As the central structure of the entire power grid,the power transformer’s defects will bring great inconvenience to people’s lives and may cause significant economic losses.And with the development of the ubiquitous electric power Internet of Things technology,the prediction of power transformer defects has become an important issue for smart grid construction.Transformer defect prediction based on the characteristics of equipment attributes and operating environment of power transformers can achieve very good results,but the imbalance of the transformer defect prediction data is a problem that has been thought to be ignored.The defect data in the power transformer data is far less than the non-defective data,which leads to the supervised learning model paying more attention to the non-defective data,which makes the model perform poorly on the defect data.In practice,we need to pay more attention to transformer defect prediction.In view of this problem,this paper first analyzes the research significance of power transformer defect prediction,and briefly introduces the research status of transformer defect prediction and the method of processing imbalanced data sets.Then use five machine learning methods and eight unbalanced data set processing methods to train and test on the processed transformer defect data set using a five-fold crossover method.The experimental results show that the tree-based integrated method Random Forest and XGBoost can predict transformer defects well,and their performance on various indicators is far superior to other models.And eight imbalanced data set processing methods can improve the recall rate of the model on defect data on a variety of supervised learning models,thereby improving the F1 value of the model on the defect data.Among them,the Adasyn imbalanced data set processing method can maintain predictions.Under the premise of accuracy and accuracy of defect data,the model’s recall rate on defect data is maximized.This paper through multi-dimensional analysis and research has confirmed that the imbalanced data processing method can improve the effectiveness of transformer defect prediction to a certain extent,and can use different imbalanced data set processing methods for different task needs,which provides a transformer defect prediction problem.New processing ideas.