Research On Transformer Fault Diagnosis Based On BP Neural Network Optimized By Cuckoo Search Algorithm

Transformer is the key apparatus for voltage grade transformation and power transmission in the power system.As a result of the fact that its operating condition has direct influence on the safety and reliability of the power grid,how to detect the latent faults immediately and accurately in the process of its operation has become a significant subject to be further studied in the power system.Based on abundant existing literature,the BP neural network optimized by cuckoo search algorithm,that is CS-BP neural network,is introduced innovatively to the area of transformer fault diagnosis in this thesis.So the accuracy and precision of transformer fault diagnosis can be effectively improved,thus providing a reliable guarantee for the proper functioning of transformer and power system.Firstly,the gas generation mechanism of transformer oil and fault classification are summarized by reading relevant materials.Then the basic principles,advantages and disadvantages of traditionally diagnostic methods as well as artificial intelligence methods are introduced.The basic mathematical principle of BP neural network is expounded through detailed formula derivation and the merits and demerits of it are analyzed.Additionally,the improved methods for BP neural network at the present stage are recommended briefly and the fundamental idea and realization process of CS algorithm are discussed as well.Secondly,the BP neural network optimized by CS algorithm is applied to the area of transformer fault diagnosis.The initial thresholds and weights of BP neural network are mapped to the nest location of CS algorithm and the optimal combination of them is determined through imitating the process in which cuckoo searches for the supreme nest,thus overcoming the disadvantages of BP neural network such as slow convergence speed and easy access to local optimum.Taking three ratios resulting from the five characteristic gas components in transformer oil as the network input and seven operating states of transformer as the network output,the CS-BP neural network is employed to delve the transformer fault information hidden in training samples and identify a certain number of test samples so as to verdict the transformer fault types.By simulating the training curve and diagnostic result based on CS-BP neural network,this thesis reaches a conclusion that the BP neural network optimized by CS algorithm can significantly decrease the training number and dramatically improve the accuracy and precision of diagnosing the test samples.In the meanwhile,by comparing the diagnostic results,it can be indicated that the CS-BP neural network shows more excellent performance for the same group of samples compared to the PSO-BP neural network.Lastly,the transformer fault diagnostic system is designed and developed by using Microsoft SQL Server 2012 software and Visual Studio 2012 software through C#and MATLAB hybrid programming technology on the basis of algorithm simulation.Three-ratio method,BP neural network,PSO-BP neural network and CS-BP neural network attached to the system are utilized comprehensively to identify the transformer fault type according to the data collected from the running scene of the transformer,then the feasibility and validity of the fault diagnostic system are verified by analyzing the transformer on the spot.