Fault Prediction Of Transformer Considering The Correlation Between State Parameters

With the development of China’s economy and the improvement of science and technology,the role of electrical energy in daily life has become more and more important,the national requirements for power supply have increased,and the safe operation of the power grid has become more important.As the core equipment in the process of energy transmission and conversion in power systems,power transformers have complex operating environments.Once they fail will cause huge economic losses,even endanger human safety,and consume a lot of human and material resources for maintenance.The transformer fault prediction technology can detect latent faults to a certain extent in advance,and staff can reasonably arrange maintenance based on the prediction results to prevent the fault from expanding,thereby avoiding power outages caused by transformer faults.Therefore,the research on transformer fault prediction methods has important practical significance for improving the reliability of power system operation.Transformer fault prediction means that after obtaining the values of the dissolved gas components in the transformer oil,predictive analysis is performed by artificial intelligence methods,future values are predicted based on historical monitoring values,and the predicted values are analyzed using transformer fault diagnosis technology to analyze what kind of operating state the transformer will be in or what kind of fault will occur.Based on this,this paper first builds a combined prediction model based on wavelet decomposition and long-short-term memory network to predict the dissolved gas concentration in transformer oil,then builds a fault diagnosis model based on the Dropout deep confidence network,and finally substitutes the predicted values into the trained fault diagnosis model to realize transformer fault prediction.The main research contents of this article are as follows:1.Analyze and summarize the current research status of transformer state parameter prediction and fault diagnosis.For most oil-immersed transformers currently in service in the power system,the gas source,gas production principle,gas analysis method and fault type judgment principle are introduced in detail.The main and secondary characteristic gases corresponding to different fault types of the transformer are summarized,and the causes of each fault are simply analyzed.On this basis,the overall model of transformer fault prediction is established,and the input characteristic parameters of the model are selected.2.Aiming at the characteristics of on-line monitoring data of transformer oil chromatography,a combined prediction method of oil dissolved gas concentration based on wavelet decomposition and long-term and short-term memory neural network was proposed.Firstly,the transformer state parameter sequence is analyzed,the association rules between the sequences are mined,and the input characteristic parameters of the prediction model are extracted.Then the wavelet transform is used to analyze the time series of the transformer oil chromatograph online to separate the low frequency trend terms and high frequency noise terms in the parameter series.Finally,the relevant parameter sequence and the decomposed sub-sequence with the prediction parameter sequence are input into each LSTM model to obtain several sets of prediction components,and the prediction components of each group are superimposed and reconstructed to obtain the final prediction result.3.Aiming at the problem that deep neural networks are susceptible to overfitting when the data is less and the model is more complicated,the Dropout algorithm is introduced to construct a DBN-Dropout fault diagnosis model.Compared with the original DBN fault diagnosis model,the network generalization ability is effectively improved.Based on common transformer fault characteristic gases,a non-coding ratio is established as the model input.The fault modes are divided into 6 types: normal,low and medium temperature overheating,high temperature and overheating,high energy discharge,low energy discharge,and partial discharge.The collected transformer fault data is proportional.Divided into training set and test set,the network parameters are determined by the training effect,and the determined model is obtained.Finally,the prediction model and diagnosis model proposed in this paper are combined to form a transformer fault prediction model.An example analysis shows that the model can predict transformer faults to a certain extent,and reasonable maintenance can be arranged according to the prediction results.