The Research Of Transformer Health Monitoring System Base On Vibration Analysis

Transformer is the key equipment for energy conversion in smart power grid.It is not only numerous,but also requires high reliability.Therefore,it is of great engineering value and scientific significance to carry out transformer fault diagnosis and health monitoring research.The common methods of transformer fault diagnosis include vibration method,oil gas analysis method,partial discharge method,recovery voltage method,frequency response analysis method and infrared diagnosis technology.Among the fault diagnosis methods of transformer,vibration method has been paid more and more attention by researchers because it can effectively diagnose potential faults of transformer in the early stage and eliminate faults as soon as possible.Based on the vibration method,this thesis extracts the features of the vibration signals on the surface of the transformer box,trains the fault diagnosis model through the features,and uses the information fusion method to analyze the output of the fault diagnosis model,and finally obtains the diagnosis results.The transformer fault diagnosis method is integrated into the transformer health monitoring system to realize the transformer health monitoring.The specific research contents of this thesis are as follows:(1)Hilbert-Huang Transform,Ensemble Empirical Mode Decomposition + Hilbert Transform,Empirical Wavelet Transform + Hilbert Transform algorithm modules are constructed,the advantages and disadvantages of the three methods for vibration feature extraction are compared and analyzed,and the Empirical Wavelet Transform + Hilbert Transform which can extract features quickly and adaptively without modal aliasing is determined.(2)BP neural network,RBF neural network,Extreme Learning Machine,Probabilistic neural network and Generalized Regression neural network were respectively established to train fault diagnosis models,and the performance of each model was evaluated by Mean Average Precision(m AP).The results show that the fault diagnosis model trained by BP neural network has better recall and accuracy.(3)Considering that a single sensor in the health monitoring system may be interfered by noise or misjudged by fault,this thesis proposes to use the multi-sensor information fusion method to improve the accuracy of transformer fault diagnosis.The fusion effects of D＿S fusion rule,Murphy fusion rule and principal component fusion rule were compared through experiments.The results show that the principal component fusion rule can avoid the problem of inaccurate judgment caused by error information more effectively.(4)Based on the fault diagnosis method studied,the real-time analysis function of transformer health monitoring system is realized by combining Matlab Production Server and Kafka.Realize the management function of massive historical data based on HDFS and Hbase;Based on Zookeeper,the high availability of cluster is realized,and an off-line computing module is developed to enable users to train fault diagnosis models for specific transformers.Finally,a transformer model of SZ11-5000/110 was monitored under the actual operating environment through the system.The experimental results show that the system can carry out real-time monitoring and fault warning on the operating parameters of the transformer.