摘要
针对微分进化DE(differential evolution)算法存在的过早出现收敛而易陷入局部最优值的不足,提出了改进DE算法与模糊聚类相结合的广义神经网络的变压器故障诊断新方法。该方法根据变压器油中的5种特征气体含量,利用自适应调整策略改进微分进化参数而进行优化模糊聚类目标函数,同时广义神经神经网络又可以发挥其训练速度快和逼近效果好等方面的优势,得出故障类型。从仿真实验结果来看,将该混合算法进行变压器故障诊断其准确率和收敛速度快都有了很好的改善,相比改进DE神经网络和FCM神经网络其逼近效果也是最好的,有利于更好的诊断故障,并通过样本验证结果进行比较。该模型简单易于实现,具有很强的实用性和泛化性。
Differential evolution algorithm exists in premature convergence easily to fall into the local optimal value shortage,the new methods of the transformer＇s fault diagnosis of generalized neural network combined medified DE algorithm and fuzzy clustering.The method adopts adaptive adjusting tactics to improve the differential evoluting parameters,thus optimizing fuzzy clustering objective function according to five characteristic gas content in transformer oil,meanuhile the generalized neural network can also give play to its fast training speed and fine appoximating effect,obtaining fault form.Seeing from the results of simulation experiment,take the mixed algorithm tomake the transformer＇s fault diagnosis,its accuracy and fast convergence speed have good imprrvement.Comparing withthe modified DE and FCM neural networks,its appoximating effect is also the best and beneficial to more diagnostic fault.Comparison by the sample result,the model is easy to be achieved and is of strong practicability and qenerulization
出处
《电气开关》
2012年第5期30-33,37,共5页
Electric Switchgear
关键词
电力变压器
改进微分进化
模糊聚类
广义神经网络
故障诊断
power transformer
improved differential evolution
fuzzy clustering
generalized neural network
fault diagnosis
