Study On Transformer Fault Diagnosis Bosed On Fuzzy Neural Network

With the fast development of social economy, people’s demand for electricity ishigher and higher, The Chinese government attaches great importance to the safety ofthe electric power industry. As one of the most important equipment for the entirepower system, transformer’s reliability related to the ability to run security andstability of the entire power system, so it has great practical significance for the faultdiagnosis of transformer.The maintenance mode of current power system equipment have transformedfrom the traditional ‘regular maintenance’ centered to scientific and reasonablereliability ‘state maintenance’, the dissolved gas analysis(DGA) technology intransformer oil is proven effective ‘state maintenance’ technology, also is a primarymethod utilized in transformer fault diagnosis.The process of transformer fault development is a nonlinear process betweenDGA data and fault type, so it is difficult to use a precise mathematical model toexpress. Neural network’s reasoning speed is very fast, and the network weightcoefficient through continuing to absorb the knowledge of self-training can also bere-optimization, so it has very strong fault tolerant, association, self-learning ability.Apply neural network in transformer fault diagnosis, without manual intervention.With its strong fault tolerance, self-learning ability to establish the nonlinear mappingof the DGA and fault type, predict the early symptoms of fault.For the nonlinear relationship between transformer DGA data and the type offault, the BP neural network fault diagnosis model is established, different algorithmare adopted to process the network and to make a comparison. The processed networkmakes a fault diagnosis and the diagnosis results are given in which variable learningrate momentum algorithm is the most accurate one with a rate of83.3percent. BPneural network has slow convergence and low precision. Its performance can not meetthe actual needs and the identification rate is difficult to rise.Fuzzy neural network is the integrated form of fuzzy logic and neural network,which, possessed structure knowledge and inferential capability of fuzzy logic, notonly remedies their singularity and deficiencies but also does apply the capability of self-study and fault-tolerant of neural network. Meanwhile, in order to furtherenhance the fault recognition rate, the paper introduce the Gauss membership functionvariable learning rate momentum method and improved gradient algorithm of fuzzyneural network, establish fuzzy neural network model of transformer fault diagnosis,and do the simulation, compared with fuzzy neural network based on the traditionalgradient method. It is suggested in the stimulation results that the optimization is veryeffective. Identifications are made to the fault diagnosis samples in which theidentification rate of the gaussian membership function variable learning ratemomentum algorithm model is91.7percent while the identification rate of theimproved gradient variable algorithm model is94.4percent.A comparative analysis of diagnosis effects is made between neural network andfuzzy neural network model. The analysis demonstrates that the fuzzy neural networkperformes better in transformer fault diagnosis due to the reason that the utilization ofgaussian membership function variable learning rate momentum algorithm and theimproved gradient variable algorithm model improves the diagnosis efficiency andprecision.Finally, the transformer is diagnosed by adopting a synthetic diagnosis approachof model diagnosis and assistive diagnosis, of which the results are in conformity withrealities.