The Research On Application Of Wavelet Network Based On GA To Transformer Fault Diagnosis

Power transformer is one of the key equipment of the power system, whose run condition impact directly on the security and stability of the power system. How to discover the incipient fault of power transformers timely and true has been the current research focus of the fault diagnosis of high voltage electric equipment. This paper first analyzed and researched the transformer faults and the diagnosis technology in the round, concluded the reason of the power transformer faults and various kinds of exceptional signs when the power transformer goes wrong or lies hidden trouble, introduced the method of the periodic check and aperiodic check of the power transformer in detail, especially the dissolved gas-in-oil analysis (DGA) technology that can distinguish the early hidden trouble of power transformer. Then analyzed synthetically the basic theory, operation mechanism and each advantage, disadvantage of genetic algorithm, artificial neural network and wavelet transform. Because of Sigmoid function characteristic, BP network is only one hypo-excellent network in fact to be improved. And BP algorithm is use to iterate by gradient decline method of error function, So its convergence speed is slow, and easily settles into local small extremum, and the concrete extremum position is closely related to the initialization of the weights. Traditional genetic algorithm is similar to the exhaustive heuristic search, though it is a global search. That causes inevitably the searching time too long. This paper has studied and designed a kind of faults diagnosis model based on GA-WANN to the concrete problem. This mode first adopted genetic algorithm to optimize the parameter of wavelet network, then adopted BP algorithm to train the network. In GA, the chromosome code adopted real number, and adopted self-adaptive probabilities of the crossover and mutation in order that the genetic algorithm not only maintains ANN characteristics, but also avoids ANN defects. Simulation result indicates this algorithm solves the problem that wavelet network settles into local small extremum so easily that the network surging will increase and the network will not be convergent if the initialization is unreasonable, and overcomes the shortcoming that the speed is too slow if use genetic algorithm to train neural network independently, at the same time the diagnosis accuracy improves to some extent too.