Research On Power Transformer Fault Diagnosis Based On Multi-parameter Information Fusion Technology

As one of the key equipments of the power system,the oil-immersed power transformer has a large fault impact and complex fault causes.In order to ensure the safety and stability of power system operation,it is necessary to monitor and judge the system operating status.With the growing maturity of the internet industry and the continuous development of sensor technology,the emergence of smart grids has resulted in intelligent on-line monitoring,condition assessment and fault diagnosis of transformers.Based on the above,this paper combines the intelligent diagnosis algorithm with the fault characteristic parameters of transformer,and studies the transformer fault diagnosis around the oil chromatographic data and electrical tests data.Firstly,based on the analysis the current status of transformer fault diagnosis and the existing problems of the applied diagnostic models,the oil chromatographic data is used as the fault characteristic parameter,PSO-ELM fusion dynamic weighted AdaBoost and PSO-IGWO optimized hybrid kernel extreme learning machine are proposed,and the simulation verification is carried out by Matlab software.However,due to the limited information of fault characteristics displayed by oil chromatographic data and the uncertainty of transformer fault diagnosis,it is impossible to give an accurate judgment result for the abnormal state of transformer operation.Therefore,this paper combines the oil chromatographic data of the transformer with the electrical tests data to form multi-feature parameters.A transformer fault diagnosis model based on multi-parameter information fusion is proposed.This model completes the initial diagnosis of transformer fault with four multi-class correlation vector machines,and transforms the diagnosis result into the corresponding evidence body.Then the DS evidence theory is improved by using the lance distance function,the spectral angle cosine function and the conflict redistribution strategy.The improved DS evidence theory is used to realize the final decision fusion of the evidence body,and the feasibility and accuracy of the model are proved by an example analysis.Secondly,the above three models,BP neural network and multi-class correlation vector machine are simulated and verified by using single oil chromatographic data and multi-feature parameters respectively.By comparing the simulation results,it is known that there is a certain gap between the diagnostic effects of different models,and the diagnostic accuracy of multi-feature parameters is significantly better than that of single oil chromatographic data.Among the three diagnostic models proposed in this paper,the transformer fault diagnosis model with multi-parameter information fusion has the highest accuracy rate of93.37%,followed by the PSO-ELM fusion dynamic weighted AdaBoost transformer faultdiagnosis,which is 89.10%.The transformer fault diagnosis model of PSO-IGWO optimized hybrid kernel extreme learning machine has the lowest,88.08%.Therefore,the multi-parameter information fusion is more suitable for determining the abnormal operating state of the transformer.Finally,the transformer fault diagnosis expert system is built using the GUI function of Matlab,which achieves good human-computer interaction,and provides convenience for better diagnosis of transformer running state.The paper has 53 figures,8 tables and 66 references.