Research On Key Technologies Of Data-Driven Intelligent Health Management For Power Transformer

As the key electrical equipment in the power grid,the safe and stable operation of power transformers is the necessary foundation for ensuring the normal supply of high quality power and the normal operation of society.With the rapid development of smart grid,the transformer state data from condition monitoring system,power production management system,operation dispatching system and environmental meteorology system have gradually possessed the typical characteristics of big data,such as volume,variety and velocity.Therefore,in the context of electric power big data,the comprehensive mining and analysis of transformer state data is of great significance for promoting refinement and intelligence level of equipment operation and maintenance.This paper takes multi-source information reflecting the operation state of equipment as the data base,and uses artificial intelligence technology and data mining technology as analysis tools to deeply study the key technologies of intelligent health management of power transformers,including anomaly detection technology,fault diagnosis technology,condition assessment technology and situation prediction technology.The main research contents and results are as follows:Aiming at the problem of abnormal data generated by the power transformer online monitoring system due to the influence of transformer operation state change,external environmental interference,communication interruption and other factors,this paper proposed a method of anomaly recognition and pattern discrimination for monitoring data.Firstly,the empirical wavelet transform(EWT)method and the autoregressive integrated moving average(ARIMA)model were used for time series modeling of monitoring data to obtain the residual sequence reflecting the anomaly monitoring data value,and then the isolation forest algorithm was further used to identify the abnormal information,and the monitoring sequence was segmented according to the recognition results.Secondly,the segmented sequence was symbolized by the improved multi-dimensional SAX vector representation method,and the judgment result of the anomaly pattern was obtained by calculating the similarity score of the adjacent symbol vectors,and the monitoring sequence correlation was further used to verify the judgment result.Finally,the case study result shows that the proposed method can reliably recognize the abnormal data and accurately distinguish between invalid and valid anomaly patterns.Aiming at the imbalanced data problem in intelligent fault diagnosis of power transformers,this paper proposed a fault data augmentation method based on deep learning to achieve the balanced distribution of the samples in different classes.Firstly,the conditional Wasserstein generative adversarial network(CWGAN)with gradient penalty was built to guide the generation of multi-category fault samples,and the training instability problem of the original generative adversarial network model was solved.Secondly,the diagnosis model based on stacked autoencoder(SAE)network was constructed by taking the non-code ratios of dissolved gases in oil as characteristic parameters,and the technical framework of power transformer fault diagnosis based on data augmentation method was further designed.Finally,the evaluation index system consisting of the accuracy,F1 score,and G-mean was selected to compare and analyze the diagnosis effects of the classifier before and after the data augmentation.The case study result indicates that the proposed fault data augmentation method can more effectively solve the classification preference problem for the majority class of the fault diagnosis model and improve its overall classification performance compared with the traditional over-sampling methods.Aiming at the problem of information uncertainty in the state evaluation of power transformers,this paper proposed a multilayer health condition assessment method for transformers considering information uncertainty.Firstly,on the basis of comprehensive consideration of the functional structure and performances of the power transformer,a multilayer condition assessment system including the equipment layer,the component layer,the defect layer and the index layer was constructed.Then,the extension cloud theory was used to solve the uncertainty problem in the state grade division.After that,the relative importance of assessment factors in each layer was measured accurately by combining analytic hierarchy process,association rules and variable weighting method.Finally,the improved Dezert-Smarandache theory(DSmT)was proposed to effectively solve the highly conflicting evidences fusion failure problem of the Dempster-Shafer(D-S)evidence theory.The verification result shows that the proposed method can accurately and effectively evaluate the health condition of the transformer and its main components,and provide detailed analysis results of component defects.Accurate prediction of development trend of gas concentration dissolved in transformer oil can provide important basises for fault early warning of power transformers.The empirical mode decomposition(EMD)method and the long short-term memory(LSTM)neural network were introduced into the prediction method of dissolved gases in oil.Firstly,the gas concentration sequence was decomposed into a group of relatively smooth components by using the EMD to reduce the mutual influence among diverse trend information.After that,forecasting models based on the LSTM neural network were constructed respectively for each subsequence,and then the Bayesian theory was used for the hyper-parameters optimization of neural networks to improve the forecasting accuracy.Finally,the prediction results of each subsequence were superimposed to obtain the final gas concentration forecasting results.The case study shows that the proposed prediction method can reflect the development trend of dissolved gas content in power transformer oil,and outperform traditional prediction algorithms with respect to forecasting accuracy.