Research On Intelligent Diagnosis Method For Transformer Fault

Transformers are complex and expensive electrical equipment in the power system.They are frequently used in the power grid.They are also one of the equipments with large capacity and high failure rate in the power grid.Once the transformer fails,it will have a significant impact on the safety and stability of the power system,Causing huge economic losses to the country and the people.Therefore,in order to avoid and reduce the losses caused by transformer faults,it is very important to carry out transformer fault diagnosis research in a timely and effective manner to find potential transformer faults.For the shortcomings of traditional neural network in transformer fault diagnosis application,network model learning speed is slow,diagnosis accuracy is low,and it is easy to fall into local extreme value,this paper studies and analyzes the transformer fault intelligent diagnosis method.The specific research contents are as follows:(1)Based on the oil-immersed power transformer,study its basic structure and common failure types.Analyze and explain the source and process of dissolved gas in transformer oil,and extract the characteristic gas components used for transformer fault diagnosis based on dissolved gas in oil.Analyze the corresponding relationship between each transformer fault type and characteristic gas to provide a basis for subsequent transformer fault diagnosis.(2)Analyze the parameter selection and realization process of genetic algorithm(GA),differential evolution algorithm(DE)and BP neural network,and explain the performance advantages and disadvantages of genetic algorithm and differential evolution algorithm by example application analysis.Based on the BP neural network,the specific implementation steps of optimizing the BP neural network using genetic algorithm and differential evolution algorithm respectively are studied to obtain the best weight and threshold parameters of the BP neural network.(3)Analyze the collected data of transformer fault samples,use normalization function to normalize the original fault sample data,and divide the fault sample data into training sample data set and test sample data set in proportion.Study the coding method of each transformer fault type,use the encoded transformer fault as the network model output vector,and use the normalized fault sample data as the network model input vector.(4)Establish BP neural network model,GA-BP network model and DE-BP network model for transformer fault diagnosis,and train the three network models with training sample set to get the final transformer fault diagnosis model;The test sample set verifies the final network model of the transformer and compares and analyzes the test results of the three network models.