Research On Fault Diagnosis Of Transformer Based On Intelligent Algorithm Optimizing Probabilistic Neural Network

As a hub node equipment connected by each voltage level in the power grid,its stable and healthy operation plays a very important role in the development of the power system.With the implementation of the construction of national comprehensive energy power grid,the power transformer changes towards higher voltage level and capacity replacement due to the development of The Times.If the transformer fails,it will cause serious harm and impact on the power system and end users.Therefore,it has practical value and practical significance to study the power transformer fault diagnosis technology and improve the level of operation and maintenance.The working principle of the transformer is relatively complex.The analysis of dissolved eas in the oil is a necessary means for fault diagnosis and latent fault monitoring.Firstly,the fault diagnosis mechanism and characteristics of dissolved gas in the oil are analyzed.In order to weaken the judgment of extreme data,a model of 3-40-9 combines the characteristics of nmproved IEC method and probability.The optimization is divided into two aspects.From the model side,the parameter values in the probability density function of the genetic algorithm are proposed,extract the smoothing parameters reflecting the whole sample space,improve the accuracy of the model diagnosis,and the training results before and after the optimization show that the diagnosis accuracy is increased from 80%to 92.5%.From the perspective of the data side,To eliminate the ambiguity of fault characteristics and fault types,Expand the dissolved gas ratio in the oil,Add the fault criteria,Using the rough set theory reduction feature to retain the core attribute closely related to the fault type as the new input feature quantity,Establish a probabilistic neural network diagnosis model of type 8-40-5-5,Remove redundancy,modest correlation of attribute information,Reduce the learning time of the fault judgment while ensuring the accuracy,Training and test simulation of the same set of fault data,Training results showed a troubleshooting accuracy improvement to 97.5%,Improve the diagnostic accuracy,high feasibility and good application prospect.