Vibration Monitoring And Predicting Methods Of Transformer Winding In Operation

The power transformer is the fundamental equipment of power gird.It plays an important role in power transmission and voltage conversion.Transformer winding fault is the majority of all transformer faults.Therefore,the real-time evaluation and fault warning of winding condition are of great significance to the stable operation of the transformer and the entire power grid.Vibration analysis is a sensitive and on-line method,which can assess the condition of winding by monitoring the transformer vibration signals.In order to improve the accuracy of vibration monitoring of transformer winding and the reliability of transformer,the characteristics of transformer in operation are firstly discussed in this paper.Then,based on the characteristics of transformer in operation,the methods of vibration monitoring and prediction are proposed.Making research on the vibration characteristics of the transformer in operation is an important premise to realize the condition monitoring of transformer winding.Based on the summary of the existing and typical methods of vibration analysis,the vibration signals of a 500 kV transformer are analyzed in time domain,frequency domain and time-frequency domain.The results show that the vibration signals of transformer in operation present non-stationary characteristics in time domain.The frequency spectrum of transformer vibration is mainly 100 Hz and its multiplications.With the variation of the load current,the vibration characteristics change complicatedly.Therefore,it is necessary to study the characteristics of transformer vibration signals carefully.In order to study the variation law between transformer vibration signals and its operation status,a method of transformer winding condition monitoring,which based on hierarchical cluster and majority voting is proposed in this paper.Firstly,the hierarchical cluster is adopted to cluster the transformer operation voltage,load current and vibration signals.Then,the three-dimensional clusters distribution of the voltage-current-vibration signals is obtained.On the basis of the distance from the monitoring signals to the fitting surface,the existence of belonging clusters and the condition of winding are judged by majority voting.The calculation results of a 500 kV transformer vibration monitoring signals show that the three-dimensional clusters distribution of the voltage-current-vibration signals can clearly reflect the vibration characteristics of the transformer in operation.The majority voting can make accurate judgment on the belonging clusters of the monitoring signals,which provides an improtant basis for the condition monitoring of transformer winding.In order to realize the fault warning of transformer winding,a method of condition prediction of transformer winding,which based on generalized regression neural network and Markov chain is proposed in this paper.Firstly,a generalized regression neural network model,which takes voltage and current as the input and vibration signals as the output is established.On the basis of the neural network model,the vibration signals are calculated.Then,the Markov chain is used to correct the relative error between the predicted and real vibration signals.The analysis of a 500 kV transformer vibration monitoring signals show that the prediction results of generalized regression neural network model modified by Markov chain has high calculation accuracy,which provides an important reference for the condition evaluation and warning of the transformer winding.The results of this paper may provide the practical value and basis for the condition monitoring and prediction of transformer winding.