Research On Multilevel Fault Diagnosis Of Power Transformer Based On DGA Analysis

Power transformer burdens on the power conversion and transmission tasks,it is an important guarantee for safe,reliable and economical power system operation.Therefore,it is of great significance for the entire system to make accurate judgments on potential failures of transformer,reduce and prevent the frequency of failures,and formulate state maintenance schedules in a timely manner.As a kind of fault diagnosis method of oil immersed power transformers with the widest range of application and the longest application time,dissolved gas analysis(DGA)has a good fault diagnosis effect.According to the characteristics and limitations of the current DGA intelligent diagnosis algorithms,the idea of multi-level grading is adopted throughout the thesis.By collecting a large amount of transformer condition data and experts' experience,this paper has further studied to map relationship between dissolved gas data and fault types to explore more efficient transformer fault diagnosis technology.The main models constructed in this paper are as follows:First of all,a fault diagnosis model based on rough set and decision fusion is constructed.After carefully analyzing the gas production mechanism of transformer faults and related references,this model is established by making some adjustments of the traditional DGA ratio,and the rough set theory are utilized to reduce the dimension of the data.Based on this,the information fusion technology is used to synthesize and analyze the multi-sourced diagnosis results from the multi-classifier to obtain more effective diagnosis conclusions.This method realizes the initial mapping of the fuzzy relationship between DGA gas and fault type.Furthermore,the limitation of single diagnosis method is overcome by the idea of multi-level machine learning and information fusion,and a complementary fusion of methods is achieved.Secondly,a multi-level fault diagnosis model of transformer based on neighborhood rough set and multi-kernel SVM is constructed.After further adjustment of the gas ratio,the neighboring rough set is used to obtain the effective feature information with high importance in each stage,and a multi-level hierarchical diagnosis model is established.In this way,the quantified information between each layer of DGA gas and the type of fault is obtained while achieving the rough set defect optimization.In addition,the multi-kernel SVM is used as a classifier to overcome the shortcomings of small kernel space and low robustness about a single kernel function.The case study proves that this method not only quantifies the fuzziness between the fault features and the fault classes,but also has a good fault judgment effect,which can further improve the diagnostic accuracy.Finally,in order to weaken the restriction of data unbalance to fault diagnosis,a fault diagnosis model based on K-S test and NSMOTEBoostSVM was constructed by using hierarchical diagnostic idea.K-S test was used to filter the effective features of each diagnostic layer in the confidence interval as fault input from the statistical point of view.On this basis,resampling mode-NSMOTE and integrated learning classifier-BoostSVM are combined to form a new classifier that can cope with data imbalance.The example shows that this method can increase the accuracy of diagnosis of a few samples,inhibit the reduction of diagnostic performance caused by the imbalance of data between classes,and have certain significance for the improvement of the overall sample diagnosis.