Research On Fault Diagnosis And State Prediction Of Transformer Based On Dissolved Gas Analysis In Oil

Transformers are crucial devices in power systems responsible for voltage regulation and power distribution,ensuring their safe and stable operation is essential for the entire power system.Dissolved Gas Analysis(DGA)technology,as an advanced and effective condition maintenance method,can monitor the concentration levels of gas components in the transformer’s insulating oil,promptly reflecting the internal operating status and potential faults of the transformer.Therefore,this study deeply investigates transformer fault diagnosis and state prediction based on DGA:(1)This paper conducts an in-depth study on the related mechanisms of dissolved gas analysis in transformer oil.This mainly includes the basic structure of transformers,online monitoring devices for dissolved gases in oil,gas generation mechanisms,and the relationship between internal transformer faults and the concentration of dissolved gases in oil,laying a solid theoretical foundation for subsequent research.(2)To address the issue of difficult performance improvement due to the scarcity of transformer fault samples and the lack of effective feature selection,this paper proposes a two-stage transformer fault diagnosis model MIFS-ASSA-SVM,which integrates a multi-filter interactive feature selection method(MIFS)and an adaptive sparrow algorithm(ASSA)optimized support vector machine(SVM).First,in MIFS,two filter methods are combined to establish a comprehensive evaluation criterion for feature importance ranking.Then,according to the sorting results,input the features and interact with the classifier dimension by dimension,dynamically selecting a subset of features with fewer dimensions and better generalization ability.At the same time,ASSA is used for SVM parameter optimization to ensure a more effective and faster optimization process.Finally,the input classifier obtains the diagnostic results based on the selected feature subset.Comparative experimental results show that the fault diagnosis model proposed in this paper performs well under various evaluation criteria,with overall diagnostic performance and individual fault recognition performance being superior and stable.(3)To address the issue of poor prediction performance of a single model due to the nonlinearity and high volatility of transformer state parameters,namely dissolved gas concentrations in oil,this paper proposes a combined dissolved gas concentration prediction model VMD-ARIMA-SVR based on Variational Mode Decomposition(VMD),Autoregressive Integrated Moving Average(ARIMA)and Support Vector Machine Regression(SVR).First,VMD is used to decompose the original gas concentration series,reducing its complexity and nonstationarity.Then,the ARIMA-SVR model,designed according to the "linear-nonlinear" framework,is used to predict the decomposed sub-series,first using ARIMA for linear prediction,and then establishing an SVR model to mine nonlinear information using the fitting residuals.Finally,the final prediction result is obtained by superimposing the sub-series prediction results.Comparative experimental results show that VMD has higher decomposition quality compared to other decomposition techniques,which is more conducive to model performance improvement.The combined model structure designed in this paper is effective and reliable,capable of capturing both linear and nonlinear information,has better prediction capabilities and robustness compared to other baseline models.In summary,this study employs advanced feature selection,model combination,and optimization techniques to develop more accurate and reliable fault diagnosis and state prediction models,which offering effective tools for industrial maintenance.