The Fundamental Frequency Amplitude Of Transformer Surface Vibration Signal Analysis And Prediction

Transformer is one of the key equipment of power system,and its operation status plays an important influence in the safe and stable operation of power grid.Transformer surface vibration signal contains a wealth of transformer status information,domestic and foreign scholars based on vibration analysis of the transformer on-line monitoring and fault diagnosis technology has done a lot of research,and made a lot of research results.The fundamental frequency(100Hz)amplitude of the transformer vibration signal is an important basis for analyzing and judging the transformer operation state and diagnosis the fault of transformer.However,due to many factors,theoretical analysis of the fundamental frequency amplitude of the transformer surface vibration signal is difficult.There is no mature method used the transformer condition monitoring based on the fundamental frequency amplitude.Aiming at the practical requirements of acquisition and analysis of transformer surface vibration signal,this paper analyzed and discussed the selection of sensor,selection of vibration measuring points and acquisition parameters,and designed and implement a portable vibration signal acquisition system what used to collect vibration data of transformer surface during running.Then,the frequency domain and energy of transformer surface vibration signal was analyzed.Combining the operating voltage and the load current data,the relationship between fundamental frequency amplitude and the data of the transformer operating was analyzed.The results showed that the fundamental frequency amplitude was affected by multiple factors.There was greatly difference between the measured value and theoretical calculation.This paper proposes a method what can be used to predict the vibration fundamental frequency amplitude of the transformer which is normal operation based on generalized regression neural network(GRNN).Neural network is trained according to historical data of operating conditions and surface vibration of the transformer.The operating data includes operating voltage,load current,oil temperature,etc.The trained network can predict the frequency amplitude of transformer surface vibration signal based on real-time operating data.Analyzing surface vibration signal on operating transformer shows that the proposed method has a higher calculation precision than the previous ways,which can provide a reference for the on-line monitoring of transformer vibration.Firstly,the feature weights are calculated based on the fuzzy entropy theory.Then the training samples are selected according to the weighted Euclidean distance of the data.The measured data showed that this method could compress the training data significantly,reduced the data redundancy,improved the network training speed and the computation speed.