Research On Transformer Fault Detection And Diagnosis

Power transformers are the core equipment in the power grid and are critical to the stable operation of the power grid.In recent years,with the rapid development of China’s economy,the urbanization process has accelerated,the electricity consumption has increased,and the installed capacity of power systems has been increasing.To ensure the reliability of power supply,it is necessary to effectively diagnose the operating state of the transformer.At present,traditional transformer diagnostic methods are prone to problems such as local optimum,deviation of results,and even inability to diagnose.Therefore,based on dissolved gas data in transformer oil,combined with intelligent algorithms to predict transformer faults,timely detection of hidden hazards or faults.Preventive measures are taken to ensure reliable operation of the grid.Taking this as a background for more indepth research has important practical significance.In addition,a single intelligent algorithm has certain defects in power transformer fault diagnosis,which can not fully meet the actual requirements.This paper proposes an improved combined RF-SVM-KNN transformer fault diagnosis algorithm.First,a preliminary determination of whether the transformer is faulty is made by combining the associated gas method with a threshold standard.Further analysis of the data,exploration of some characteristic gas laws,combined with expert experience,obtained 22 feature quantities and selected 15 important feature quantities.Then,the random forest(RF),support vector machine(SVM)and K-nearest neighbor algorithm(KNN)are studied.The grid search is used to optimize the parameters,and the optimal parameters are obtained.The base classifier model for transformer fault diagnosis is established.The model is validated and analyzed.Finally,the three base classifier models are merged and complemented by weighted voting method,and the improved algorithm model is obtained.The model is evaluated and verified,and compared with the three-ratio method,RF,SVM,KNN and others’ improved methods.With case analysis,it is fully verified that the fusion RF-SVM-KNN model can effectively identify the fault type,and has the advantages of high classification accuracy,fast speed and strong generalization ability.This paper has 32 pictures,42 tables and 90 references.