Electrical Fault Diagnosis And Residual Life Prediction Technology Of Power Transformer

The state condition of power transformers directly affects the reliable operation of the power system.Up to now,a large number of power transformers in China have been in service for nearly 30 years.Equipment aging and other electrical problems are increasingly serious.Therefore,it is of great theoretical and practical significance for the reliable operation of power transformers to collect the historical fault data,strengthen the monitoring of the operation status,timely detect and deal with the potential equipment faults under the on-load operation status,and prevent and reduce the probability of power transformer faults.The transformer diagnosis method is gradually changing from the off-line monitoring of dissolved gas analysis(DGA)in oil mainly based on preventive test to the on-line monitoring of transformer faults which mainly include comprehensive diagnosis.Among all kinds of faults of power transformer,the electrical faults mainly include winding,core,insulation bushing and tap switch are the most and the most important.And most of the fault manifestations are overheat and discharge.This paper first introduces the importance of transformer diagnosis research and exposition the development status of transformer fault diagnosis.Base on the analyzing of electrical fault characteristic parameters,the method of dividing the types of power transformer electrical fault by mixing the dissolved gas in oil with the electrical fault characteristic parameters is proposed.Considering power transformer historical data for no labels,we set a based electric fault diagnosis model of power transformer base on the Deep Belief Networks(DBN),the model uses the multi-layer Restricted boltzmann machine(RBM)stack and on the top floor return parameter with BP neural Network(BPNN),the output data use SOFTMAX label for fault classification and diagnosis.The simulation results show that the fault diagnosis model of power transformer based on DBN has better convergence speed and diagnosis accuracy than the traditional BPNN model.On this basis,according to the power transformer running state obey the Markov property,set up a on-condition maintenance based on hidden Markov model(HMM)which obey the failure rate model(Cox model),Predicted the residual using life(RUL)of power transformer under the condition of partial observable data.Compared with the traditionalmethod,the proposed algorithm has a higher prediction accuracy which provides a new reference method for apparent maintenance of power transformers.