广东电力2025，Vol.38Issue(6)：68-78,11.DOI:10.3969/j.issn.1007-290X.2025.06.008
基于混合特征选择和IBSLO-KELM的变压器故障诊断方法
Transformer Fault Diagnosis Method Based on Hybrid Feature Selection and IBSLO-KELM
摘要
Abstract
In order to improve the quality of dissolved gas features in transformer oil and the accuracy of model diagnosis,this paper proposes a transformer fault diagnosis method based on hybrid feature selection and improved blood-sucking leech optimizer(IBSLO)to optimize kernel extreme learning machine(KELM)for unbalanced samples.Firstly,the original sample data is expanded by extending the natural neighborhood oversampling algorithm to achieve fault sample equalization.Secondly,a 30-dimensional candidate feature set is constructed based on the correlation ratio method,and then an hybrid feature selection method is used to integrate the rankings generated by four different feature selection methods to form a global integrated feature ranking through the rank aggregation algorithm,and a dimension-by-dimension diagnostic is used to obtain the preferred feature set.Afterwards,a good point set strategy,inverse proximity dyadic learning strategy and multiplication and division strategy are introduced to improve the optimization algorithm of blood-sucking leech,and the improved algorithm is used to optimize the relevant parameters of KELM so as to improve the classification ability of KELM.Finally,comparison experiments are conducted on different feature selection methods and different fault diagnosis models.The experimental results show that after sample expansion and feature optimization,the diagnostic accuracy of the IBSLO-KELM model can reach 97.8%,which is improved by 7.2%,5.0%,8.9%and 8.4%compared with the four base feature selection methods,proving the effectiveness of the method.关键词
不均衡样本/混合特征选择/改进吸血水蛭优化/核极限学习机/变压器故障诊断Key words
unbalanced sample/hybrid feature selection/improved blood-sucking leech optimizer/kernel extreme learning machine/transformer fault diagnosis分类
信息技术与安全科学引用本文复制引用
李海龙,杜江..基于混合特征选择和IBSLO-KELM的变压器故障诊断方法[J].广东电力,2025,38(6):68-78,11.基金项目
国家自然科学基金项目(52007047) （52007047）
天津市自然科学基金重点项目(19JCZDJC32100) （19JCZDJC32100）
