基于SMOTE-XGBoost的变压器缺陷预测

A Transformer Defect Prediction Model Based on SMOTE-XGBoost

变压器在电力系统中发挥着变换电压、分配电能等重要作用。其故障缺陷将严重危及电力系统的安全运行,从而造成巨大的经济损失和不良的社会影响。因此,变压器的缺陷故障诊断对电力系统的安全运行和社会的发展至关重要。现有关于变压器缺陷诊断的研究颇多,但变压器缺陷数据集类别不平衡仍是准确预测变压器缺陷的难点。为了解决这一难题,提出了一种基于SMOTE(Synthetic Minority Oversampling Technique)-XGBoost(Extreme Gradient Boosting)的变压器缺陷预测模型。首先,通过SMOTE算法来处理不平衡数据集,然后构建XGBoost模型预测变压器缺陷。为了表现模型的优越性,研究增加了对比实验。即将原始数据集和经过Up_Sample,CSL和Down_Sample与SMOTE四种方法处理后的数据集分别与决策树(CART)、支持向量机(SVM)、Logistic回归和XGBoost模型四种预测模型两两组合进行预测。实验结果表明,基于SMOTE-XGBoost模型的变压器缺陷预测效果最优。

The transformer is important to the power system,and its fault and defects will seriously endanger the safe operation of the power system,resulting in huge economic losses and adverse social impact.Its fault diagnosis is key to system operation and social development.Although there are many related studies,the imbalance classification of transformer defect data set is still unsolved in accurately predicting transformer health status.Therefore,the paper presents a transformer defect prediction model based on SMOTE(synthetic minority oversampling technique)-XGBoost(Extreme Gradient Boosting).First,the SMOTE(Synthetic Minority Oversampling Technique)algorithm is used to balance the data set,and then the integrated model XGBoost is used as the classification model.The comparative experiments are added to show the superiority of the model.That is,the original data sets and processed data sets processed by Up_Sample CSL,Down_Sample,and SMOTE are used to predict with the CART,SVM,Logistic regression,and XGBoost model respectively.Experimental results show that the SMOTE-XGBoost model improves transformer defect prediction accuracy.