Transformer Fault Diagnosis by Combination of Rough Set and Neural Network
Jingming Zhang, Xiao Qianhua, Shisheng Wang · Gao dianya jishu · 2007
Rough Set is a tool to extract knowledge from a vast volume of data.BP neural network has nonlinear characteristic and the ability of self organization and self-learning.A new method combining Rough Set and BP neural network is used for transformer fault diagnosis.Rough Set is just suitable for discrete data,parts of continuous attributes of decision table are discretizated based on transformer DGA knowledge,and parts of continuous attributes of decision table are discretizated with Na ve algorithm and Equal Frequency Intervals algorithm.The decision table is reduced according to the Rough Set theory and the minimal diagnostic table rules are gotten.It reflects relation of five-ratio of dissolved gases and transformer fault diagnosis,and it is a kind of improvement of the IEC three-ratio transformer diagnosis method.BP neural network has three layers: one input layer,one hidden layer and one output layer,nodes of input layer is five,nodes of hidden layer is twenty,nodes of output layer is five.The BP neural network is trained by the minimal diagnostic table,the actual malfunction diagnostic ability is proved and fault diagnosis accuracy is higher compared with IEC three-ratio diagnosis.Combining Rough Set theory with neural network,the method is applied in the transformer fault diagnosis,it reduces transformer decision table of fault diagnosis,simplifies the structure of neural network,improves the diagnostic accuracy.