Transformer Winding Deformation Diagnosis Based On Feature Extraction And ISSA-PNN

Transformers operate in a complex env ironment with many structural components,and various faults will inevitably occur during operation,among which winding deformation occurs more frequently and has the greatest impact on transformer operation.In this thesis,the magnetic induction intensity of the leakage field of transformer is used as a characteristic to characterize the winding deformation state,and the winding deformation diagnosis of transformer is realized by analyzing the nonlinear mapping relationship between the winding deformati on type and the leakage magnetic characteristics.To address the problem that the leakage features cannot fully chara cterize the transformer winding state due to unsound a priori knowledge,a leakage feature extraction method based on feature evaluation an d Kernel Principal Component Analysis is studied.The redundant features are eliminated by feature evaluation,and the key features reflecting the winding deformation state are extracted by KPCA.It is shown that the feature extraction method based on feature evaluation and KPCA can reduce the correlation between the leakage features and express more key winding state information with fewer features.In order to deal with the complex nonlinear mapping relationship between transformer winding deformation types and leakage characteristics and to achieve effective discrimination of winding states,a probabilistic neura l network(PNN)based transformer winding deformation diagnosis method is studied.Meanwhile,the Sparrow Search Algorithm(SSA)is applied to the smoothing factor selection of PNN to address the problem that the smoothing factor setting of PNN is too depe ndent on the operator’s experience.By analyzing the fault information contained in the samples,the optimal smoothing factor based on the actual sample distribution is selected,a nd the feasibility of SSA-PNN for winding deformation diagnosis is verified by case analysis.An Improved Sparrow Search Algorithm(ISSA)is studied for the problem of search stagnation and falling into local optim um during SSA iteration,and it is used for transformer winding deformation diagnosis.By optimizing the population initialization process of ISSA,the quality of the initial individuals of the population is improved.The location information difference be tween population individuals is incorporated as an adjustment factor in the location update method of ISSA,and the search performance of ISSA is improved by strengthening the location information linkage between population individuals.The algorithm analy sis shows that the proposed method coordinates the global search performance and local pioneering ability of ISSA,and can further improve the diagnosis accuracy of transformer winding deformation.