Study On IFA-PNN And Improved Nuclear Fisher Algorithm In Fault Diagnosis Of Power Transformer

As a part of grid,transformer plays the role of transforming and transmitting electric energy.With the development of smart power grid,it is of great significance for the safe and stable operation of power grid to detect potential faults of transformers as early as possible.In addition,in order to reduce the power failure and excessive maintenance cost caused by regular maintenance,online intelligent monitoring is also necessary.Based on the above background and the current research status,this paper,based on the artificial intelligence method,makes the following research on transformer fault diagnosis:In view of the problems in transformer fault diagnosis,such as the inability of DGA algorithm to fully reflect fault types and the complex structure of some Neural network models,this paper proposes the IFA-PNN(improved Firefly algorithm-Probabilistic Neural Networks)diagnosis model.First,the improved firefly algorithm is used to find the best smooth factor(sigma)in the probabilistic neural network.In this model,the fusion ratio method is adopted to process the input data to make it more representative,which can ensure the accuracy of PNN neural network diagnosis.The improved firefly algorithm guarantees the global searching ability and improves the searching precision.It can find the best value quickly and accurately.In addition,the probabilistic neural network has simple structure and high classification accuracy.The experimental results show that this method has high accuracy in fault diagnosis.The transformer fault data is discrete and nonlinear,so an improved nuclear Fisher diagnosis model is proposed.In this model,Euclidean distance is used to weight the inter-class matrix,and the compound kernel function is used to replace the single kernel function.The improved model reduces the overlap of fault data in projection and makes it self-adaptive.The composite kernel function has the characteristics of multiple kernel functions and strong nonlinear processing ability,which is especially suitable for transformer data with large amount and strong nonlinear.The test shows that this method is an effective fault diagnosis method.There are 35 figures,11 tables and 72 references.