针对变压器故障样本类别不平衡造成分类模型准确率偏低的问题，提出一种基于样本集成学习和蛇优化算法（SO）优化支持向量机（SVM）的变压器故障诊断模型。该模型先利用EasyEnsemble采样器对样本进行多次欠采样后生成类别平衡的多个子集；然后以Bagging策略训练SO优化关键参数后的SVM模型，综合各个分类器结果得到最终故障类型。通过算例对所提模型有效性进行验证，数据表明，SO-SVM的故障诊断相比于RF、SVM、KNN等模型，诊断准确率分别提高了3.44%、6.89%、10.92%，AUC值分别提高了0.026 4、0.042 5、0.081 2；在同一分类器下，SO-SVM模型相比于SMOTE和ADASYN样本平衡方法，诊断准确率分别提高了4.59%、2.87%，说明SO-SVM模型对不平衡样本的故障诊断能力更优。

Aiming at the problem of low accuracy of classification model caused by the unbalanced category of transformer fault samples, a transformer fault diagnosis model based on sample integration learning and snake optimisation algorithm (SO) optimised support vector machine (SVM) is proposed. The model first uses the EasyEnsemble sampler to generate multiple subsets with balanced categories after multiple under-sampling of the samples; then the SVM model optimised by SO with key parameters is trained with the Bagging strategy, and the final fault types are obtained by integrating the results of the classifiers. The validity of the proposed model is verified by the arithmetic example, and the data show that the diagnostic accuracy of SO-SVM′s fault diagnosis is improved by 3.44%, 6.89%, 10.92%, and the AUC value is improved by 0.026 4, 0.042 5, and 0.081 2, respectively, compared with the models of RF, SVM and KNN; in the same classifier, the SO-SVM model is more accurate than the SMOTE and ADASYN sample balancing methods, the diagnostic accuracy is improved by 4.59% and 2.87%, respectively, indicating that the SO-SVM model has better fault diagnosis capability for unbalanced samples.

云南省教育厅科学研究基金资助项目(2022J1279)；云南电网有限责任公司科技项目(YNKJXM20180736)