基于DGA特征量优选与改进磷虾群算法优化支持向量机的变压器故障诊断模型

Fault diagnosis model of power transformers based on DGA features selection and improved krill herd algorithm optimization SVM

针对单一的特征气体或特征气体比值作为DGA特征量无法全面反映变压器故障分类的问题,文章从混合DGA特征量中优选出一组DGA新特征组合作为输入,建立改进磷虾群(Improved Krill Herd,IKH)算法优化支持向量机(Support Vector Machine,SVM)的变压器故障诊断模型对变压器进行故障诊断。将SVM的c和σ与11种候选特征量进行二进制编码,利用遗传算法结合支持向量机对DGA特征量进行优选,得到一组DGA新特征组合;利用IKH算法对SVM的参数进行优化,同时结合交叉验证原理构建IKH算法优化SVM的变压器故障诊断模型。基于IEC TC 10的诊断结果表明:与DGA全数据、三比值特征量相比,新DGA特征组合的故障诊断准确率分别高出10.14%和30.2%;IKHSVM准确率也要高于标准SVM和GASVM(分别为73.87%、81.13%和86.27%),说明该方法能有效诊断变压器故障。

In order to make up a single feature gas or feature ratios as DGA feature cannot fully reflect the transformer fault,a DGA feature set generate by the hybrid DGA feature set are used as the input vectors,and a transformer fault diagnosis model based on the improved krill herd(IKH)optimization support vector machine(SVM)is proposed to diagnose transformer fault.Firstly,the SVM parameters and 11 features are encoded by a binary code technique.Then,a DGA feature set for fault diagnosis of power transformers is selected by genetic algorithm(GA)and SVM.Finally,the SVM parameters are optimized by GA,and a transformer fault diagnosis model is conducted based on IKH algorithm optimization SVM combined with the cross validation principle.The fault diagnosis results based on IEC TC 10 show that the proposed DGA feature set increases the accuracy by 10.14%and 30.2%over the DGA data and IEC ratios.Furthermore,the accuracy rate of IKHSVM is better than the standard SVM and GASVM(accuracy rate are 73.87%,81.13%and 86.27%),which verifies the accuracy of transformer fault diagnosis can be improved by the proposed method.